KERNEL PROFILING GUIDE

v2023.1.1 | March 2023

User Manual
 TABLE OF CONTENTS

Chapter 1. Introduction.........................................................................................1
1.1. Profiling Applications.....................................................................................1
Chapter 2. Metric Collection...................................................................................3
2.1. Sets and Sections......................................................................................... 3
2.2. Sections and Rules........................................................................................4
2.3. Replay.......................................................................................................5
2.3.1. Kernel Replay........................................................................................ 5
2.3.2. Application Replay.................................................................................. 6
2.3.3. Range Replay......................................................................................... 8
2.3.3.1. Defining Ranges................................................................................ 8
2.3.3.2. Supported APIs................................................................................. 9
2.3.4. Application Range Replay.........................................................................13
2.3.5. Graph Profiling..................................................................................... 13
2.4. Compatibility............................................................................................. 14
2.5. Profile Series............................................................................................. 14
2.6. Overhead..................................................................................................15
Chapter 3. Metrics Guide..................................................................................... 17
3.1. Hardware Model......................................................................................... 17
3.2. Metrics Structure........................................................................................ 22
3.3. Metrics Decoder......................................................................................... 27
3.4. Range and Precision.................................................................................... 31
Chapter 4. Metrics Reference................................................................................33
Chapter 5. Sampling............................................................................................45
5.1. Warp Scheduler States................................................................................. 45
Chapter 6. Reproducibility....................................................................................46
6.1. Serialization.............................................................................................. 46
6.2. Clock Control.............................................................................................46
6.3. Cache Control............................................................................................ 47
6.4. Persistence Mode........................................................................................ 48
Chapter 7. Special Configurations...........................................................................49
7.1. Multi Instance GPU......................................................................................49
Chapter 8. Roofline Charts................................................................................... 51
8.1. Overview.................................................................................................. 51
8.2. Analysis....................................................................................................52
Chapter 9. Memory Chart..................................................................................... 54
9.1. Overview.................................................................................................. 54
Chapter 10. Memory Tables.................................................................................. 57
10.1. Shared Memory......................................................................................... 57
10.2. L1/TEX Cache...........................................................................................59
10.3. L2 Cache................................................................................................ 62

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | ii
 10.4. L2 Cache Eviction Policies............................................................................65
10.5. Device Memory......................................................................................... 66
Chapter 11. FAQ................................................................................................ 68

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | iii
 LIST OF TABLES

Table 1 Available Sections ...................................................................................... 4

Table 2 Replay modes and metrics per GPU workload type ............................................. 14

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | iv
Chapter 1.
INTRODUCTION

This guide describes various profiling topics related to NVIDIA Nsight Compute and
NVIDIA Nsight Compute CLI. Most of these apply to both the UI and the CLI version of
the tool.
To use the tools effectively, it is recommended to read this guide, as well as at least the
following chapters of the CUDA Programming Guide:
‣ Programming Model
‣ Hardware Implementation
‣ Performance Guidelines
Afterwards, it should be enough to read the Quickstart chapter of the NVIDIA Nsight
Compute or NVIDIA Nsight Compute CLI documentation, respectively, to start using
the tools.

1.1. Profiling Applications

During regular execution, a CUDA application process will be launched by the user.
It communicates directly with the CUDA user-mode driver, and potentially with the
CUDA runtime library.

When profiling an application with NVIDIA Nsight Compute, the behavior is different.
The user launches the NVIDIA Nsight Compute frontend (either the UI or the CLI)

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 1
 Introduction

on the host system, which in turn starts the actual application as a new process on the
target system. While host and target are often the same machine, the target can also be a
remote system with a potentially different operating system.
The tool inserts its measurement libraries into the application process, which allow
the profiler to intercept communication with the CUDA user-mode driver. In addition,
when a kernel launch is detected, the libraries can collect the requested performance
metrics from the GPU. The results are then transferred back to the frontend.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 2
Chapter 2.
METRIC COLLECTION

Collection of performance metrics is the key feature of NVIDIA Nsight Compute. Since
there is a huge list of metrics available, it is often easier to use some of the tool's pre-
defined sets or sections to collect a commonly used subset. Users are free to adjust which
metrics are collected for which kernels as needed, but it is important to keep in mind the
Overhead associated with data collection.

2.1. Sets and Sections

NVIDIA Nsight Compute uses Section Sets (short sets) to decide, on a very high level,
the amount of metrics to be collected. Each set includes one or more Sections, with each
section specifying several logically associated metrics. For example, one section might
include only high-level SM and memory utilization metrics, while another could include
metrics associated with the memory units, or the HW scheduler.
The number and type of metrics specified by a section has significant impact on the
overhead during profiling. To allow you to quickly choose between a fast, less detailed
profile and a slower, more comprehensive analysis, you can select the respective section
set. See Overhead for more information on profiling overhead.
By default, a relatively small number of metrics is collected. Those mostly include high-
level utilization information as well as static launch and occupancy data. The latter two
are regularly available without replaying the kernel launch. The default set is collected
when no --set, --section and no --metrics options are passed on the command
line. The full set of sections can be collected with --set full.
Use --list-sets to see the list of currently available sets. Use --list-sections to
see the list of currently available sections. The default search directory and the location
of pre-defined section files are also called sections/. All related command line options
can be found in the NVIDIA Nsight Compute CLI documentation.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 3
 Metric Collection

2.2. Sections and Rules

Table 1 Available Sections

Identifier and Filename Description

ComputeWorkloadAnalysis Detailed analysis of the compute resources of the streaming
(Compute Workload Analysis) multiprocessors (SM), including the achieved instructions per clock
(IPC) and the utilization of each available pipeline. Pipelines with
very high utilization might limit the overall performance.
InstructionStats (Instruction Statistics of the executed low-level assembly instructions (SASS).
Statistics) The instruction mix provides insight into the types and frequency of
the executed instructions. A narrow mix of instruction types implies
a dependency on few instruction pipelines, while others remain
unused. Using multiple pipelines allows hiding latencies and enables
parallel execution.
LaunchStats (Launch Statistics) Summary of the configuration used to launch the kernel. The launch
configuration defines the size of the kernel grid, the division of
the grid into blocks, and the GPU resources needed to execute the
kernel. Choosing an efficient launch configuration maximizes device
utilization.
MemoryWorkloadAnalysis Detailed analysis of the memory resources of the GPU. Memory can
(Memory Workload Analysis) become a limiting factor for the overall kernel performance when
fully utilizing the involved hardware units (Mem Busy), exhausting
the available communication bandwidth between those units (Max
Bandwidth), or by reaching the maximum throughput of issuing
memory instructions (Mem Pipes Busy). Depending on the limiting
factor, the memory chart and tables allow to identify the exact
bottleneck in the memory system.
NUMA Affinity (NumaAffinity) Non-uniform memory access (NUMA) affinities based on compute and
memory distances for all GPUs.
Nvlink (Nvlink) High-level summary of NVLink utilization. It shows the total received
and transmitted (sent) memory, as well as the overall link peak
utilization.
Nvlink_Tables (Nvlink_Tables) Detailed tables with properties for each NVLink.
Nvlink_Topology NVLink Topology diagram shows logical NVLink connections with
(Nvlink_Topology) transmit/receive throughput.
Occupancy (Occupancy) Occupancy is the ratio of the number of active warps per
multiprocessor to the maximum number of possible active warps.
Another way to view occupancy is the percentage of the hardware's
ability to process warps that is actively in use. Higher occupancy
does not always result in higher performance, however, low
occupancy always reduces the ability to hide latencies, resulting in
overall performance degradation. Large discrepancies between the
theoretical and the achieved occupancy during execution typically
indicates highly imbalanced workloads.
SchedulerStats (Scheduler Summary of the activity of the schedulers issuing instructions. Each
Statistics) scheduler maintains a pool of warps that it can issue instructions
for. The upper bound of warps in the pool (Theoretical Warps) is
limited by the launch configuration. On every cycle each scheduler

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 4
 Metric Collection

Identifier and Filename Description

checks the state of the allocated warps in the pool (Active Warps).
Active warps that are not stalled (Eligible Warps) are ready to issue
their next instruction. From the set of eligible warps, the scheduler
selects a single warp from which to issue one or more instructions
(Issued Warp). On cycles with no eligible warps, the issue slot is
skipped and no instruction is issued. Having many skipped issue slots
indicates poor latency hiding.
SourceCounters (Source Source metrics, including branch efficiency and sampled warp stall
Counters) reasons. Warp Stall Sampling metrics are periodically sampled over
the kernel runtime. They indicate when warps were stalled and
couldn't be scheduled. See the documentation for a description of
all stall reasons. Only focus on stalls if the schedulers fail to issue
every cycle.
SpeedOfLight (GPU Speed Of High-level overview of the throughput for compute and memory
Light Throughput) resources of the GPU. For each unit, the throughput reports the
achieved percentage of utilization with respect to the theoretical
maximum. Breakdowns show the throughput for each individual
sub-metric of Compute and Memory to clearly identify the highest
contributor.
WarpStateStats (Warp State Analysis of the states in which all warps spent cycles during the
Statistics) kernel execution. The warp states describe a warp's readiness or
inability to issue its next instruction. The warp cycles per instruction
define the latency between two consecutive instructions. The
higher the value, the more warp parallelism is required to hide this
latency. For each warp state, the chart shows the average number
of cycles spent in that state per issued instruction. Stalls are not
always impacting the overall performance nor are they completely
avoidable. Only focus on stall reasons if the schedulers fail to issue
every cycle.

2.3. Replay
Depending on which metrics are to be collected, kernels might need to be replayed one
or more times, since not all metrics can be collected in a single pass. For example, the
number of metrics originating from hardware (HW) performance counters that the
GPU can collect at the same time is limited. In addition, patch-based software (SW)
performance counters can have a high impact on kernel runtime and would skew results
for HW counters.

2.3.1. Kernel Replay

In Kernel Replay, all metrics requested for a specific kernel instance in NVIDIA Nsight
Compute are grouped into one or more passes. For the first pass, all GPU memory that
can be accessed by the kernel is saved. After the first pass, the subset of memory that is
written by the kernel is determined. Before each pass (except the first one), this subset is
restored in its original location to have the kernel access the same memory contents in
each replay pass.
NVIDIA Nsight Compute attempts to use the fastest available storage location for
this save-and-restore strategy. For example, if data is allocated in device memory, and

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 5
 Metric Collection

there is still enough device memory available, it is stored there directly. If it runs out
of device memory, the data is transferred to the CPU host memory. Likewise, if an
allocation originates from CPU host memory, the tool first attempts to save it into the
same memory location, if possible.
As explained in Overhead, the time needed for this increases the more memory is
accessed, especially written, by a kernel. If NVIDIA Nsight Compute determines that
only a single replay pass is necessary to collect the requested metrics, no save-and-
restore is performed at all to reduce overhead.

2.3.2. Application Replay

In Application Replay, all metrics requested for a specific kernel launch in NVIDIA
Nsight Compute are grouped into one or more passes. In contrast to Kernel Replay, the
complete application is run multiple times, so that in each run one of those passes can be
collected per kernel.
For correctly identifying and combining performance counters collected from multiple
application replay passes of a single kernel launch into one result, the application needs
to be deterministic with respect to its kernel activities and their assignment to GPUs,
contexts, streams, and potentially NVTX ranges. Normally, this also implies that the
application needs to be deterministic with respect to its overall execution.
Application replay has the benefit that memory accessed by the kernel does not need to
be saved and restored via the tool, as each kernel launch executes only once during the

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 6
 Metric Collection

lifetime of the application process. Besides avoiding memory save-and-restore overhead,

application replay also allows to disable Cache Control. This is especially useful if other
GPU activities preceding a specific kernel launch are used by the application to set
caches to some expected state.
In addition, application replay can support profiling kernels that have
interdependencies to the host during execution. With kernel replay, this class of kernels
typically hangs when being profiled, because the necessary responses from the host
are missing in all but the first pass. In contrast, application replay ensures the correct
behavior of the program execution in each pass.
In contrast to kernel replay, multiple passes collected via application replay imply
that all host-side activities of the application are duplicated, too. If the application
requires significant time for e.g. setup or file-system access, the overhead will increase
accordingly.

Across application replay passes, NVIDIA Nsight Compute matches metric data for the
individual, selected kernel launches. The matching strategy can be selected using the --
app-replay-match option. For matching, only kernels within the same process and
running on the same device are considered. By default, the grid strategy is used, which
matches launches according to their kernel name and grid size. When multiple launches
have the same attributes (e.g. name and grid size), they are matched in execution order.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 7
 Metric Collection

2.3.3. Range Replay

In Range Replay, all requested metrics in NVIDIA Nsight Compute are grouped into
one or more passes. In contrast to Kernel Replay and Application Replay, Range Replay
captures and replays complete ranges of CUDA API calls and kernel launches within
the profiled application. Metrics are then not associated with individual kernels but
with the entire range. This allows the tool to execute kernels without serialization and
thereby supports profiling kernels that should be run concurrently for correctness or
performance reasons.

2.3.3.1. Defining Ranges

Range replay requires you to specify the range for profiling in the application. A range
is defined by a start and an end marker and includes all CUDA API calls and kernels
launched between these markers from any CPU thread. The application is responsible
for inserting appropriate synchronization between threads to ensure that the anticipated
set of API calls is captured. Range markers can be set using one of the following options:
‣ Profiler Start/Stop API
Set the start marker using cu(da)ProfilerStart and the end marker using
cu(da)ProfilerStop. Note: The CUDA driver API variants of this API require
to include cudaProfiler.h. The CUDA runtime variants require to include
cuda_profiler_api.h.
This is the default for NVIDIA Nsight Compute.
‣ NVTX Ranges
Define the range using an NVTX Include expression. The range capture starts with
the first CUDA API call and ends at the last API call for which the expression is
matched, respectively. If multiple expressions are specified, a range is defined
as soon as any of them matches. Hence, multiple expressions can be used to
conveniently capture and profile multiple ranges for the same application execution.
The application must have been instrumented with the NVTX API for any
expressions to match.
This mode is enabled by passing --nvtx --nvtx-include <expression> [--
nvtx-include <expression>] to the NVIDIA Nsight Compute CLI.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 8
 Metric Collection

Ranges must fulfill several requirements:

‣ It must be possible to synchronize all active CUDA contexts at the start of the range.
‣ Ranges must not include unsupported CUDA API calls. See Supported APIs for the
list of currently supported APIs.
In addition, there are several recommendations that ranges should comply with to
guarantee a correct capture and replay:
‣ Set ranges as narrow as possible for capturing a specific set of CUDA kernel
lanuches. The more API calls are included, the higher the potentially created
overhead from capturing and replaying these API calls.
‣ Avoid freeing host allocations written by device memory during the range. This
includes both heap as well as stack allocations. NVIDIA Nsight Compute does
not intercept creation or destruction of generic host (CPU)-based allocations.
However, to guarantee correct program execution after any replay of the range,
the tool attempts to restore host allocations that were written from device memory
during the capture. If these host addresses are invalid or re-assigned, the program
behavior is undefined and potentially unstable. In cases where avoiding freeing
such allocations is not possible, you should limit profiling to one range using --
launch-count 1, set the disable-host-restore range replay option and optionally use
--kill yes to terminate the process after this range.
‣ When defining the range markers using cu(da)ProfilerStart/Stop, prefer
the CUDA driver API calls cuProfilerStart/Stop. Internally, NVIDIA Nsight
Compute only intercepts the CUDA driver API variants and the CUDA runtime API
may not trigger these if no CUDA context is active on the calling thread.

2.3.3.2. Supported APIs

Range replay supports a subset of the CUDA API for capture and replay. This page lists
the supported functions as well as any further, API-specific limitations that may apply.
If an unsupported API call is detected in the captured range, an error is reported and the
range cannot be profiled. The groups listed below match the ones found in the CUDA
Driver API documentation.
Generally, range replay only captures and replay CUDA Driver API calls. CUDA
Runtime APIs calls can be captured when they generate only supported CUDA Driver
API calls internally. Deprecated APIs are not supported.

Error Handling
All supported.

Initialization
Not supported.

Version Management
All supported.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 9
 Metric Collection

Device Management
All supported, except:

‣ cuDeviceSetMemPool

Primary Context Management

‣ cuDevicePrimaryCtxGetState

Context Management
All supported, except:

‣ cuCtxSetCacheConfig
‣ cuCtxSetSharedMemConfig

Module Management
‣ cuModuleGetFunction
‣ cuModuleGetGlobal
‣ cuModuleGetSurfRef
‣ cuModuleGetTexRef
‣ cuModuleLoad
‣ cuModuleLoadData
‣ cuModuleLoadDataEx
‣ cuModuleLoadFatBinary
‣ cuModuleUnload

Library Management
All supported, except:

‣ cuKernelSetAttribute
‣ cuKernelSetCacheConfig

Memory Management
‣ cuArray*
‣ cuDeviceGetByPCIBusId
‣ cuDeviceGetPCIBusId
‣ cuMemAlloc
‣ cuMemAllocHost
‣ cuMemAllocPitch
‣ cuMemcpy*
‣ cuMemFree

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 10
 Metric Collection

‣ cuMemFreeHost
‣ cuMemGetAddressRange
‣ cuMemGetInfo
‣ cuMemHostAlloc
‣ cuMemHostGetDevicePointer
‣ cuMemHostGetFlags
‣ cuMemHostRegister
‣ cuMemHostUnregister
‣ cuMemset*
‣ cuMipmapped*

Virtual Memory Management

Not supported.

Stream Ordered Memory Allocator

Not supported.

Unified Addressing
Not supported.

Stream Management
‣ cuStreamCreate*
‣ cuStreamDestroy
‣ cuStreamGet*
‣ cuStreamQuery
‣ cuStreamSetAttribute
‣ cuStreamSynchronize
‣ cuStreamWaitEvent

Event Management
All supported.

External Resource interoperability

Not supported.

Stream Memory Operations

Not supported.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 11
 Metric Collection

Execution Control
‣ cuFuncGetAttribute
‣ cuFuncGetModule
‣ cuFuncSetAttribute
‣ cuFuncSetCacheConfig
‣ cuLaunchCooperativeKernel
‣ cuLaunchHostFunc
‣ cuLaunchKernel

Graph Management
Not supported.

Occupancy
All supported.

Texture/Surface Reference Management

Not supported.

Texture Object Management

All supported.

Surface Object Management

All supported.

Peer Context Memory Access

Not supported.

Graphics Interoperability
Not supported.

Driver Entry Point Access

All supported.

Surface Object Management

All supported.

OpenGL Interoperability
Not supported.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 12
 Metric Collection

VDPAU Interoperability
Not supported.

EGL Interoperability
Not supported.

2.3.4. Application Range Replay

In Application Range Replay, all requested metrics in NVIDIA Nsight Compute are
grouped into one or more passes. Similar to Range Replay, metrics are not associated
with individual kernels but with the entire selected range. This allows the tool to execute
workloads (kernels, CUDA graphs, ...) without serialization and thereby supports
profiling workloads that must be run concurrently for correctness or performance
reasons.
In contrast to Range Replay, the range is not explicitly captured and executed directly
for each pass, but instead the entire application is re-run multiple times, with one pass
collected for each range in every application execution. This has the benefit that no
application state must be observed and captured for each range and API calls within the
range do not need to be supported explicitly, as correct execution of the range is handled
by the application itself.
Defining ranges to profile is identical to Range Replay. The CUDA context for which the
range should be profiled must be current to the thread defining the start of the range
and must be active for the entire range.

2.3.5. Graph Profiling

In multiple replay modes, NVIDIA Nsight Compute can profile CUDA graphs as single
workload entities, rather than profile individual kernel nodes. The behavior can be
toggled in the respective command line or UI options.
The primary use cases for enabling this mode are:
‣ Profile graphs that include mandatory concurrent kernel nodes.
‣ Profile graphs that include device-sided graph launches.
‣ Profile graph behavior more accurately across multiple kernel node launches, as
caches are not purged in between nodes.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 13
 Metric Collection

Note that when graph profiling is enabled, certain metrics such as instruction-level
source metrics are not available. This then also applies to kernels profiled outside of
graphs.

2.4. Compatibility
The set of available replay modes and metrics depends on the type of GPU workload to
profile.

Table 2 Replay modes and metrics per GPU workload type

Workload Replay Mode Metric Groups

Type
Kernel Application Range Hardware Unit- Instruction- Launch
Counters / Level Level
SMSP Source Source
Kernel Yes Yes Yes Yes Yes Yes Yes
Range No No Yes Yes Yes No Some
Cmdlist Yes No No Yes Yes No Some
Graph 1 Yes No No Yes Yes No Some

2.5. Profile Series

The performance of a kernel is highly dependent on the used launch parameters. Small
changes to the launch parameters can have a significant effect on the runtime behavior
of the kernel. However, identifying the best parameter set for a kernel by manually
testing a lot of combinations can be a tedious process.
To make this workflow faster and more convenient, Profile Series provide the ability
to automatically profile a single kernel multiple times with changing parameters. The
parameters to be modified and values to be tested can be independently enabled and
configured. For each combination of selected parameter values a unique profile result
is collected. And the modified parameter values are tracked in the description of the
results of a series. By comparing the results of a profile series, the kernel’s behavior
on the changing parameters can be seen and the most optimal parameter set can be
identified quickly.

1
Limitations also apply to kernels profiled outside of graphs.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 14
 Metric Collection

2.6. Overhead
As with most measurements, collecting performance data using NVIDIA Nsight
Compute CLI incurs some runtime overhead on the application. The overhead does
depend on a number of different factors:
‣ Number and type of collected metrics
Depending on the selected metric, data is collected either through a hardware
performance monitor on the GPU, through software patching of the kernel
instructions or via a launch or device attribute. The overhead between these
mechanisms varies greatly, with launch and device attributes being "statically"
available and requiring no kernel runtime overhead.
Furthermore, only a limited number of metrics can be collected in a single pass of
the kernel execution. If more metrics are requested, the kernel launch is replayed
multiple times, with its accessible memory being saved and restored between
subsequent passes to guarantee deterministic execution. Therefore, collecting more

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 15
 Metric Collection

metrics can significantly increase overhead by requiring more replay passes and
increasing the total amount of memory that needs to be restored during replay.
‣ The collected section set
Since each set specifies a group of section to be collected, choosing a less
comprehensive set can reduce profiling overhead. See the --set command in the
NVIDIA Nsight Compute CLI documentation.
‣ Number of collected sections
Since each section specifies a set metrics to be collected, selecting fewer sections can
reduce profiling overhead. See the --section command in the NVIDIA Nsight
Compute CLI documentation.
‣ Number of profiled kernels
By default, all selected metrics are collected for all launched kernels. To reduce the
impact on the application, you can try to limit performance data collection to as few
kernel functions and instances as makes sense for your analysis. See the filtering
commands in the NVIDIA Nsight Compute CLI documentation.
There is a relatively high one-time overhead for the first profiled kernel in each
context to generate the metric configuration. This overhead does not occur for
subsequent kernels in the same context, if the list of collected metrics remains
unchanged.
‣ GPU Architecture
For some metrics, the overhead can vary depending on the exact chip they are
collected on, e.g. due to varying number of units on the chip. Similarly, the overhead
for resetting the L2 cache in-between kernel replay passes depends on the size of
that cache.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 16
Chapter 3.
METRICS GUIDE

3.1. Hardware Model

Compute Model
All NVIDIA GPUs are designed to support a general purpose heterogeneous parallel
programming model, commonly known as Compute. This model decouples the GPU
from the traditional graphics pipeline and exposes it as a general purpose parallel multi-
processor. A heterogeneous computing model implies the existence of a host and a
device, which in this case are the CPU and GPU, respectively. At a high level view, the
host (CPU) manages resources between itself and the device and will send work off to
the device to be executed in parallel.
Central to the compute model is the Grid, Block, Thread hierarchy, which defines how
compute work is organized on the GPU. The hierarchy from top to bottom is as follows:

‣ A Grid is a 1D, 2D or 3D array of thread blocks.

‣ A Block is a 1D, 2D or 3D array of threads, also known as a Cooperative Thread Array
(CTA).
‣ A Thread is a single thread which runs on one of the GPU's SM units.
The purpose of the Grid, Block, Thread hierarchy is to expose a notion of locality
amongst a group of threads, i.e. a Cooperative Thread Array (CTA). In CUDA, CTAs
are referred to as Thread Blocks. The architecture can exploit this locality by providing
fast shared memory and barriers between the threads within a single CTA. When a
Grid is launched, the architecture guarantees that all threads within a CTA will run
concurrently on the same SM. Information on the grids and blocks can be found in the
Launch Statistics section.
The number of CTAs that fit on each SM depends on the physical resources required by
the CTA. These resource limiters include the number of threads and registers, shared
memory utilization, and hardware barriers. The number CTAs per SM is referred to as

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 17
 Metrics Guide

the CTA occupancy, and these physical resources limit this occupancy. Details on the
kernel's occupancy are collected by the Occupancy section.
Each CTA can be scheduled on any of the available SMs, where there is no guarantee in
the order of execution. As such, CTAs must be entirely independent, which means it is
not possible for one CTA to wait on the result of another CTA. As CTAs are independent,
the host (CPU) can launch a large Grid that will not fit on the hardware all at once,
however any GPU will still be able to run it and produce the correct results.
CTAs are further divided into groups of 32 threads called Warps. If the number of
threads in a CTA is not dividable by 32, the last warp will contain the remaining number
of threads.
The total number of CTAs that can run concurrently on a given GPU is referred to as
Wave. Consequently, the size of a Wave scales with the number of available SMs of a
GPU, but also with the occupancy of the kernel.

Streaming Multiprocessor
The Streaming Multiprocessor (SM) is the core processing unit in the GPU. The SM is
optimized for a wide diversity of workloads, including general-purpose computations,
deep learning, ray tracing, as well as lighting and shading. The SM is designed to
simultaneously execute multiple CTAs. CTAs can be from different grid launches.
The SM implements an execution model called Single Instruction Multiple Threads
(SIMT), which allows individual threads to have unique control flow while still
executing as part of a warp. The Turing SM inherits the Volta SM's independent thread
scheduling model. The SM maintains execution state per thread, including a program
counter (PC) and call stack. The independent thread scheduling allows the GPU to
yield execution of any thread, either to make better use of execution resources or to
allow a thread to wait for data produced by another thread possibly in the same warp.
Collecting the Source Counters section allows you to inspect instruction execution and
predication details on the Source Page, along with Sampling information.
Each SM is partitioned into four processing blocks, called SM sub partitions. The SM sub
partitions are the primary processing elements on the SM. Each sub partition contains
the following units:

‣ Warp Scheduler
‣ Register File
‣ Execution Units/Pipelines/Cores

‣ Integer Execution units

‣ Floating Point Execution units
‣ Memory Load/Store units
‣ Special Function unit
‣ Tensor Cores

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 18
 Metrics Guide

Shared within an SM across the four SM partitions are:

‣ Unified L1 Data Cache / Shared Memory

‣ Texture units
‣ RT Cores, if available
A warp is allocated to a sub partition and resides on the sub partition from launch
to completion. A warp is referred to as active or resident when it is mapped to a sub
partition. A sub partition manages a fixed size pool of warps. On Volta architectures,
the size of the pool is 16 warps. On Turing architectures the size of the pool is 8 warps.
Active warps can be in eligible state if the warp is ready to issue an instruction. This
requires the warp to have a decoded instruction, all input dependencies resolved, and
for the function unit to be available. Statistics on active, eligible and issuing warps can be
collected with the Scheduler Statistics section.
A warp is stalled when the warp is waiting on

‣ an instruction fetch,
‣ a memory dependency (result of memory instruction),
‣ an execution dependency (result of previous instruction), or
‣ a synchronization barrier.
See Warp Scheduler States for the list of stall reasons that can be profiled and the Warp
State Statistics section for a summary of warp states found in the kernel execution.
The most important resource under the compiler's control is the number of registers
used by a kernel. Each sub partition has a set of 32-bit registers, which are allocated by
the HW in fixed-size chunks. The Launch Statistics section shows the kernel's register
usage.

Memory
Global memory is a 49-bit virtual address space that is mapped to physical memory
on the device, pinned system memory, or peer memory. Global memory is visible to all
threads in the GPU. Global memory is accessed through the SM L1 and GPU L2.
Local memory is private storage for an executing thread and is not visible outside of that
thread. It is intended for thread-local data like thread stacks and register spills. Local
memory addresses are translated to global virtual addresses by the the AGU unit. Local
memory has the same latency as global memory. One difference between global and
local memory is that local memory is arranged such that consecutive 32-bit words are
accessed by consecutive thread IDs. Accesses are therefore fully coalesced as long as all
threads in a warp access the same relative address (e.g., same index in an array variable,
same member in a structure variable, etc.).
Shared memory is located on chip, so it has much higher bandwidth and much lower
latency than either local or global memory. Shared memory can be shared across a
compute CTA. Compute CTAs attempting to share data across threads via shared

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 19
 Metrics Guide

memory must use synchronization operations (such as __syncthreads()) between stores

and loads to ensure data written by any one thread is visible to other threads in the
CTA. Similarly, threads that need to share data via global memory must use a more
heavyweight global memory barrier.
Shared memory has 32 banks that are organized such that successive 32-bit words map
to successive banks that can be accessed simultaneously. Any 32-bit memory read or
write request made of 32 addresses that fall in 32 distinct memory banks can therefore
be serviced simultaneously, yielding an overall bandwidth that is 32 times as high as the
bandwidth of a single request. However, if two addresses of a memory request fall in the
same memory bank, there is a bank conflict and the access has to be serialized.
A shared memory request for a warp does not generate a bank conflict between two
threads that access any address within the same 32-bit word (even though the two
addresses fall in the same bank). When multiple threads make the same read access,
one thread receives the data and then broadcasts it to the other threads. When multiple
threads write to the same location, only one thread succeeds in the write; which thread
that succeeds is undefined.
Detailed memory metrics are collected by the Memory Workload Analysis section.

Caches
All GPU units communicate to main memory through the Level 2 cache, also known
as the L2. The L2 cache sits between on-chip memory clients and the framebuffer. L2
works in physical-address space. In addition to providing caching functionality, L2 also
includes hardware to perform compression and global atomics.

The Level 1 Data Cache, or L1, plays a key role in handling global, local, shared, texture,
and surface memory reads and writes, as well as reduction and atomic operations. On
Volta and Turing architectures there are , there are two L1 caches per TPC, one for each
SM. For more information on how L1 fits into the texturing pipeline, see the TEX unit
description. Also note that while this section often uses the name "L1", it should be
understood that the L1 data cache, shared data, and the Texture data cache are one and
the same.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 20
 Metrics Guide

L1 receives requests from two units: the SM and TEX. L1 receives global and local
memory requests from the SM and receives texture and surface requests from TEX.
These operations access memory in the global memory space, which L1 sends through a
secondary cache, the L2.
Cache hit and miss rates as well as data transfers are reported in the Memory Workload
Analysis section.

Texture/Surface
The TEX unit performs texture fetching and filtering. Beyond plain texture memory
access, TEX is responsible for the addressing, LOD, wrap, filter, and format conversion
operations necessary to convert a texture read request into a result.
TEX receives two general categories of requests from the SM via its input interface:
texture requests and surface load/store operations. Texture and surface memory space
resides in device memory and are cached in L1. Texture and surface memory are
allocated as block-linear surfaces (e.g. 2D, 2D Array, 3D). Such surfaces provide a cache-
friendly layout of data such that neighboring points on a 2D surface are also located
close to each other in memory, which improves access locality. Surface accesses are
bounds-checked by the TEX unit prior to accessing memory, which can be used for
implementing different texture wrapping modes.
The L1 cache is optimized for 2D spatial locality, so threads of the same warp that read
texture or surface addresses that are close together in 2D space will achieve optimal
performance. The L1 cache is also designed for streaming fetches with constant latency;
a cache hit reduces DRAM bandwidth demand but not fetch latency. Reading device
memory through texture or surface memory presents some benefits that can make it an
advantageous alternative to reading memory from global or constant memory.
Information on texture and surface memory can be found in the Memory Workload
Analysis section.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 21
 Metrics Guide

3.2. Metrics Structure

Metrics Overview
NVIDIA Nsight Compute uses an advanced metrics calculation system, designed to
help you determine what happened (counters and metrics), and how close the program
reached to peak GPU performance (throughputs as a percentage). Every counter
has associated peak rates in the database, to allow computing its throughput as a
percentage.
Throughput metrics return the maximum percentage value of their constituent counters.
These constituents have been carefully selected to represent the sections of the GPU
pipeline that govern peak performance. While all counters can be converted to a %-of-
peak, not all counters are suitable for peak-performance analysis; examples of unsuitable
counters include qualified subsets of activity, and workload residency counters. Using
throughput metrics ensures meaningful and actionable analysis.
Two types of peak rates are available for every counter: burst and sustained. Burst rate
is the maximum rate reportable in a single clock cycle. Sustained rate is the maximum
rate achievable over an infinitely long measurement period, for "typical" operations.
For many counters, burst equals sustained. Since the burst rate cannot be exceeded,
percentages of burst rate will always be less than 100%. Percentages of sustained rate can
occasionally exceed 100% in edge cases.

Metrics Entities
While in NVIDIA Nsight Compute, all performance counters are named metrics, they
can be split further into groups with specific properties. For metrics collected via the
PerfWorks measurement library, the following entities exist:
Counters may be either a raw counter from the GPU, or a calculated counter value.
Every counter has four sub-metrics under it, which are also called roll-ups:

.sum The sum of counter values across all unit

instances.
.avg The average counter value across all unit
instances.
.min The minimum counter value across all unit
instances.
.max The maximum counter value across all
unit instances.

Counter roll-ups have the following calculated quantities as built-in sub-metrics:

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 22
 Metrics Guide

.peak_sustained the peak sustained rate

.peak_sustained_active the peak sustained rate during unit active
cycles
.peak_sustained_active.per_secondthe peak sustained rate during unit active
cycles, per second *
.peak_sustained_elapsed the peak sustained rate during unit
elapsed cycles
the peak sustained rate during unit
.peak_sustained_elapsed.per_second
elapsed cycles, per second *
.peak_sustained_region the peak sustained rate over a user-
specified "range"
.peak_sustained_region.per_secondthe peak sustained rate over a user-
specified "range", per second *
.peak_sustained_frame the peak sustained rate over a user-
specified "frame"
.peak_sustained_frame.per_second the peak sustained rate over a user-
specified "frame", per second *
.per_second the number of operations per second
.per_cycle_active the number of operations per unit active
cycle
.per_cycle_elapsed the number of operations per unit elapsed
cycle
.per_cycle_in_region the number of operations per user-
specified "range" cycle
.per_cycle_in_frame the number of operations per user-
specified "frame" cycle
.pct_of_peak_sustained_active % of peak sustained rate achieved during
unit active cycles
.pct_of_peak_sustained_elapsed % of peak sustained rate achieved during
unit elapsed cycles
.pct_of_peak_sustained_region % of peak sustained rate achieved over a
user-specified "range"
.pct_of_peak_sustained_frame % of peak sustained rate achieved over a
user-specified "frame"

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 23
 Metrics Guide

* sub-metrics added in NVIDIA Nsight Compute 2022.2.0.

Example: ncu --query-metrics-mode suffix --metrics sm__inst_executed
--chip ga100

Ratios have three sub-metrics:

.pct The value expressed as a percentage.

.ratio The value expressed as a ratio.
.max_rate The ratio's maximum value.

Example: ncu --query-metrics-mode suffix --metrics

smsp__average_warp_latency --chip ga100

Throughputs indicate how close a portion of the GPU reached to peak rate. Every
throughput has the following sub-metrics:

.pct_of_peak_sustained_active % of peak sustained rate achieved during

unit active cycles
.pct_of_peak_sustained_elapsed % of peak sustained rate achieved during
unit elapsed cycles
.pct_of_peak_sustained_region % of peak sustained rate achieved over a
user-specified "range" time
.pct_of_peak_sustained_frame % of peak sustained rate achieved over a
user-specified "frame" time

Example: ncu --query-metrics-mode suffix --metrics sm__throughput --

chip ga100

Throughputs have a breakdown of underlying metrics from which the throughput

value is computed. You can collect breakdown:<throughput-metric> to collect a
throughput's breakdown metrics.
Deprecated counter sub-metrics: The following sub-metrics were removed in NVIDIA
Nsight Compute 2022.2.0, due to not being useful for performance optimization:

.peak_burst the peak burst rate

.pct_of_peak_burst_active % of peak burst rate achieved during unit
active cycles
.pct_of_peak_burst_elapsed % of peak burst rate achieved during unit
elapsed cycles
.pct_of_peak_burst_region % of peak burst rate achieved over a user-
specified "range"

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 24
 Metrics Guide

.pct_of_peak_burst_frame % of peak burst rate achieved over a user-

specified "frame"

Deprecated throughput sub-metrics: The following sub-metrics were removed in

NVIDIA Nsight Compute 2022.2, due to not being useful for performance optimization:

.pct_of_peak_burst_active % of peak burst rate achieved during unit

active cycles
.pct_of_peak_burst_elapsed % of peak burst rate achieved during unit
elapsed cycles
.pct_of_peak_burst_region % of peak burst rate achieved over a user-
specified "range" time
.pct_of_peak_burst_frame % of peak burst rate achieved over a user-
specified "frame" time

In addition to PerfWorks metrics, NVIDIA Nsight Compute uses several other

measurement providers that each generate their own metrics. These are explained in the
Metrics Reference.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 25
 Metrics Guide

Metrics Examples

## non-metric names -- *not* directly evaluable

sm__inst_executed # counter
smsp__average_warp_latency # ratio
sm__throughput # throughput

## a counter's four first-level sub-metrics -- all evaluable

sm__inst_executed.sum
sm__inst_executed.avg
sm__inst_executed.min
sm__inst_executed.max

## all names below are metrics -- all evaluable

l1tex__data_bank_conflicts_pipe_lsu.sum
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_active
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_active.per_second
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_elapsed
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_elapsed.per_second
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_frame
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_frame.per_second
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_region
l1tex__data_bank_conflicts_pipe_lsu.sum.peak_sustained_region.per_second
l1tex__data_bank_conflicts_pipe_lsu.sum.per_cycle_active
l1tex__data_bank_conflicts_pipe_lsu.sum.per_cycle_elapsed
l1tex__data_bank_conflicts_pipe_lsu.sum.per_cycle_in_frame
l1tex__data_bank_conflicts_pipe_lsu.sum.per_cycle_in_region
l1tex__data_bank_conflicts_pipe_lsu.sum.per_second
l1tex__data_bank_conflicts_pipe_lsu.sum.pct_of_peak_sustained_active
l1tex__data_bank_conflicts_pipe_lsu.sum.pct_of_peak_sustained_elapsed
l1tex__data_bank_conflicts_pipe_lsu.sum.pct_of_peak_sustained_frame
l1tex__data_bank_conflicts_pipe_lsu.sum.pct_of_peak_sustained_region
...

Metrics Naming Conventions

Counters and metrics _generally_ obey the naming scheme:

‣ Unit-Level Counter :
unit__(subunit?)_(pipestage?)_quantity_(qualifiers?)
‣ Interface Counter :
unit__(subunit?)_(pipestage?)_(interface)_quantity_(qualifiers?)
‣ Unit Metric : (counter_name).(rollup_metric)
‣ Sub-Metric : (counter_name).(rollup_metric).(submetric)
where

‣ unit: A logical or physical unit of the GPU

‣ subunit: The subunit within the unit where the counter was measured. Sometimes
this is a pipeline mode instead.
‣ pipestage: The pipeline stage within the subunit where the counter was measured.
‣ quantity: What is being measured. Generally matches the dimensional units.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 26
 Metrics Guide

‣ qualifiers: Any additional predicates or filters applied to the counter. Often, an

unqualified counter can be broken down into several qualified sub-components.
‣ interface: Of the form sender2receiver, where sender is the source-unit and
receiver is the destination-unit.
‣ rollup_metric: One of sum, avg, min, max.
‣ submetric: refer to section Metrics Entities
Components are not always present. Most top-level counters have no qualifiers. Subunit
and pipestage may be absent where irrelevant, or there may be many subunit specifiers
for detailed counters.

Cycle Metrics
Counters using the term cycles in the name report the number of cycles in the unit's
clock domain. Unit-level cycle metrics include:

‣ unit__cycles_elapsed : The number of cycles within a range. The cycles'

DimUnits are specific to the unit's clock domain.
‣ unit__cycles_active : The number of cycles where the unit was processing
data.
‣ unit__cycles_stalled : The number of cycles where the unit was unable to
process new data because its output interface was blocked.
‣ unit__cycles_idle : The number of cycles where the unit was idle.
Interface-level cycle counters are often (not always) available in the following variations:

‣ unit__(interface)_active : Cycles where data was transferred from source-

unit to destination-unit.
‣ unit__(interface)_stalled : Cycles where the source-unit had data, but the
destination-unit was unable to accept data.

3.3. Metrics Decoder

The following explains terms found in NVIDIA Nsight Compute metric names, as
introduced in Metrics Structure.

Units
dram Device (main) memory, where the GPUs global and local memory resides.

fbpa The FrameBuffer Partition is a memory controller which sits between the level 2

cache (LTC) and the DRAM. The number of FBPAs varies across GPUs.

fe The Frontend unit is responsible for the overall flow of workloads sent by the

driver. FE also facilitates a number of synchronization operations.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 27
 Metrics Guide

gpc The General Processing Cluster contains SM, Texture and L1 in the form of TPC(s). It

is replicated several times across a chip.

gpu The entire Graphics Processing Unit.

gr Graphics Engine is responsible for all 2D and 3D graphics, compute work, and

synchronous graphics copying work.

idc The InDexed Constant Cache is a subunit of the SM responsible for caching

constants that are indexed with a register.

l1tex The Level 1 (L1)/Texture Cache is located within the GPC. It can be used as

directed-mapped shared memory and/or store global, local and texture data in its

cache portion. l1tex__t refers to its Tag stage. l1tex__m refers to its Miss stage.

l1tex__d refers to its Data stage.

ltc The Level 2 cache.

ltcfabric The LTC fabric is the communication fabric for the L2 cache partitions.

lts A Level 2 (L2) Cache Slice is a sub-partition of the Level 2 cache. lts__t refers to its

Tag stage. lts__m refers to its Miss stage. lts__d refers to its Data stage.

sm The Streaming Multiprocessor handles execution of a kernel as groups of 32

threads, called warps. Warps are further grouped into cooperative thread arrays

(CTA), called blocks in CUDA. All warps of a CTA execute on the same SM. CTAs

share various resources across their threads, e.g. the shared memory.

smsp Each SM is partitioned into four processing blocks, called SM sub partitions. The

SM sub partitions are the primary processing elements on the SM. A sub partition

manages a fixed size pool of warps.

sys Logical grouping of several units.

tpc Thread Processing Clusters are units in the GPC. They contain one or more SM,

Texture and L1 units, the Instruction Cache (ICC) and the Indexed Constant Cache

(IDC).

Subunits
aperture_device Memory interface to local device memory (dram)

aperture_peer Memory interface to remote device memory

aperture_sysmem Memory interface to system memory

global Global memory is a 49-bit virtual address space that is mapped to physical memory

on the device, pinned system memory, or peer memory. Global memory is visible to

all threads in the GPU. Global memory is accessed through the SM L1 and GPU L2.

lg Local/Global memory

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 28
 Metrics Guide

local Local memory is private storage for an executing thread and is not visible outside

of that thread. It is intended for thread-local data like thread stacks and register

spills. Local memory has the same latency as global memory.

lsu Load/Store unit

lsuin Load/Store input

mio Memory input/output

mioc Memory input/output control

shared Shared memory is located on chip, so it has much higher bandwidth and much

lower latency than either local or global memory. Shared memory can be shared

across a compute CTA.

surface Surface memory

texin TEXIN

texture Texture memory

xbar The Crossbar (XBAR) is responsible for carrying packets from a given source unit to

a specific destination unit.

Pipelines
adu Address Divergence Unit. The ADU is responsible for address divergence handling

for branches/jumps. It also provides support for constant loads and block-level

barrier instructions.

alu Arithmetic Logic Unit. The ALU is responsible for execution of most bit

manipulation and logic instructions. It also executes integer instructions, excluding

IMAD and IMUL. On NVIDIA Ampere architecture chips, the ALU pipeline performs

fast FP32-to-FP16 conversion.

cbu Convergence Barrier Unit. The CBU is responsible for warp-level convergence,

barrier, and branch instructions.

fma Fused Multiply Add/Accumulate. The FMA pipeline processes most FP32 arithmetic

(FADD, FMUL, FMAD). It also performs integer multiplication operations (IMUL,

IMAD), as well as integer dot products. On GA10x, FMA is a logical pipeline that

indicates peak FP32 and FP16x2 performance. It is composed of the FMAHeavy and

FMALite physical pipelines.

fmaheavy Fused Multiply Add/Accumulate Heavy. FMAHeavy performs FP32 arithmetic (FADD,

FMUL, FMAD), FP16 arithmetic (HADD2, HMUL2, HFMA2), and integer dot products.

fmalite Fused Multiply Add/Accumulate Lite. FMALite performs FP32 arithmetic (FADD,

FMUL, FMA) and FP16 arithmetic (HADD2, HMUL2, HFMA2).

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 29
 Metrics Guide

fp16 Half-precision floating-point. On Volta, Turing and NVIDIA GA100, the FP16 pipeline

performs paired FP16 instructions (FP16x2). It also contains a fast FP32-to-FP16 and

FP16-to-FP32 converter. Starting with GA10x chips, this functionality is part of the

FMA pipeline.

fp64 Double-precision floating-point. The implementation of FP64 varies greatly per

chip.

lsu Load Store Unit. The LSU pipeline issues load, store, atomic, and reduction

instructions to the L1TEX unit for global, local, and shared memory. It also issues

special register reads (S2R), shuffles, and CTA-level arrive/wait barrier instructions

to the L1TEX unit.

tex Texture Unit. The SM texture pipeline forwards texture and surface instructions

to the L1TEX unit's TEXIN stage. On GPUs where FP64 or Tensor pipelines are

decoupled, the texture pipeline forwards those types of instructions, too.

tma Tensor Memory Access Unit. Provides efficient data transfer mechanisms

between global and shared memories with the ability to understand and traverse

multidimensional data layouts.

uniform Uniform Data Path. This scalar unit executes instructions where all threads use the

same input and generate the same output.

xu Transcendental and Data Type Conversion Unit. The XU pipeline is responsible for

special functions such as sin, cos, and reciprocal square root. It is also responsible

for int-to-float, and float-to-int type conversions.

Quantities
instruction An assembly (SASS) instruction. Each executed instruction may generate zero or

more requests.

request A command into a HW unit to perform some action, e.g. load data from some

memory location. Each request accesses one or more sectors.

sector Aligned 32 byte-chunk of memory in a cache line or device memory. An L1 or L2

cache line is four sectors, i.e. 128 bytes. Sector accesses are classified as hits if the

tag is present and the sector-data is present within the cache line. Tag-misses and

tag-hit-data-misses are all classified as misses.

tag Unique key to a cache line. A request may look up multiple tags, if the thread

addresses do not all fall within a single cache line-aligned region. The L1 and L2

both have 128 byte cache lines. Tag accesses may be classified as hits or misses.

wavefront Unique "work package" generated at the end of the processing stage for requests.

All work items of a wavefront are processed in parallel, while work items of

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 30
 Metrics Guide

different wavefronts are serialized and processed on different cycles. At least one

wavefront is generated for each request.

A simplified model for the processing in L1TEX for Volta and newer architectures can
be described as follows: When an SM executes a global or local memory instruction for
a warp, a single request is sent to L1TEX. This request communicates the information for
all participating threads of this warp (up to 32). For local and global memory, based on
the access pattern and the participating threads, the request requires to access a number
of cache lines, and sectors within these cache lines. The L1TEX unit has internally
multiple processing stages operating in a pipeline.
A wavefront is the maximum unit that can pass through that pipeline stage per cycle. If
not all cache lines or sectors can be accessed in a single wavefront, multiple wavefronts
are created and sent for processing one by one, i.e. in a serialized manner. Limitations
of the work within a wavefront may include the need for a consistent memory space, a
maximum number of cache lines that can be accessed, as well as various other reasons.
Each wavefront then flows through the L1TEX pipeline and fetches the sectors handled
in that wavefront. The given relationships of the three key values in this model are
requests:sectors is 1:N, wavefronts:sectors 1:N, and requests:wavefronts is 1:N.
A wavefront is described as a (work) package that can be processed at once, i.e. there is a
notion of processing one wavefront per cycle in L1TEX. Wavefronts therefore represent
the number of cycles required to process the requests, while the number of sectors per
request is a property of the access pattern of the memory instruction for all participating
threads. For example, it is possible to have a memory instruction that requires 4 sectors
per request in 1 wavefront. However, you can also have a memory instruction having 4
sectors per request, but requiring 2 or more wavefronts.

3.4. Range and Precision

Overview
In general, measurement values that lie outside the expected logical range of a metric
can be attributed to one or more of the below root-causes. If values are exceeding such
range, they are not clamped by the tool to their expected value on purpose to ensure that
the rest of the profiler report remains self-consistent.

Asynchronous GPU activity

GPU engines other than the one measured by a metric (display, copy engine, video
encoder, video decoder, etc.) potentially access shared resources during profiling. Such
chip-global shared resources include L2, DRAM, PCIe, and NVLINK. If the kernel
launch is small, the other engine(s) can cause significant confusion in e.g. the DRAM
results, since it is not possible to isolate the DRAM traffic of the SM. To reduce the

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 31
 Metrics Guide

impact of such asynchronous units, consider profiling on a GPU without active display
and without other processes that can access the GPU at the time.

Multi-pass data collection

Out-of-range metrics often occur when the profiler replays the kernel launch to collect
metrics, and work distribution is significantly different across replay passes. A metric
such as hit rate (hits / queries) can have significant error if hits and queries are collected
on different passes and the kernel does not saturate the GPU to reach a steady state
(generally > 20 µs). Similarly, it can show unexpected values when the workload is
inherently variable, as e.g. in the case of spin loops.
To mitigate the issue, when applicable try to increase the measured workload to allow
the GPU to reach a steady state for each launch. Reducing the number of metrics
collected at the same time can also improve precision by increasing the likelihood that
counters contributing to one metric are collected in a single pass.

Tool issue
If you still observe metric issues after following the guidelines above, please reach out to
us and describe your issue.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 32
Chapter 4.
METRICS REFERENCE

Overview
Most metrics in NVIDIA Nsight Compute can be queried using the ncu command line
interface's --query-metrics option.
The following metrics can be collected explicitly, but are not listed by --query-
metrics, and do not follow the naming scheme explained in Metrics Structure. They
should be used as-is instead.
launch__* metrics are collected per kernel launch, and do not require an additional
replay pass. They are available as part of the kernel launch parameters (such as grid size,
block size, ...) or are computed using the CUDA Occupancy Calculator.

Launch Metrics
launch__block_dim_x Number of threads for the kernel launch in X dimension.

launch__block_dim_y Number of threads for the kernel launch in Y dimension.

launch__block_dim_z Number of threads for the kernel launch in Z dimension.

launch__block_size Total number of threads per block for the kernel launch.

launch__cluster_dim_x Number of clusters for the kernel launch in X dimension.

launch__cluster_dim_y Number of clusters for the kernel launch in Y dimension.

launch__cluster_dim_z Number of clusters for the kernel launch in Z dimension.

launch__cluster_max_active Maximum number of clusters that can co-exist on the target device. The runtime

environment may affect how the hardware schedules the clusters, so the calculated

occupancy is not guaranteed to be achievable.

launch__cluster_max_potential_size Largest valid cluster size for the kernel function and launch configuration.

launch__cluster_scheduling_policy Cluster scheduling policy.

launch__context_id CUDA context id for the kernel launch.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 33
 Metrics Reference

launch__device_id CUDA device id for the kernel launch.

launch__func_cache_config On devices where the L1 cache and shared memory use the same hardware

resources, this is the preferred cache configuration for the CUDA function. The

runtime will use the requested configuration if possible, but it is free to choose a

different configuration if required.

launch__function_pcs Kernel function entry PCs.

launch__graph_contains_device_launch Set to 1 if any node in the profiled graph can launch a CUDA device graph.

launch__graph_is_device_launchable Set to 1 if the profiled graph was device-launchable.

launch__grid_dim_x Number of blocks for the kernel launch in X dimension.

launch__grid_dim_y Number of blocks for the kernel launch in Y dimension.

launch__grid_dim_z Number of blocks for the kernel launch in Z dimension.

launch__grid_size Total number of blocks for the kernel launch.

launch__occupancy_cluster_gpu_pct Overall GPU occupancy due to clusters.

launch__occupancy_cluster_pct The ratio of active blocks to the max possible active blocks due to clusters.

launch__occupancy_limit_blocks Occupancy limit due to maximum number of blocks managable per SM.

launch__occupancy_limit_registers Occupancy limit due to register usage.

launch__occupancy_limit_shared_mem Occupancy limit due to shared memory usage.

launch__occupancy_limit_warps Occupancy limit due to block size.

launch__occupancy_per_block_size Number of active warps for given block size.

Instance values map from number of warps (uint64) to value (uint64).

launch__occupancy_per_cluster_size Number of active clusters for given cluster size.

Instance values map from number of clusters (uint64) to value (uint64).

launch__occupancy_per_register_count Number of active warps for given register count.

Instance values map from number of warps (uint64) to value (uint64).

launch__occupancy_per_shared_mem_size Number of active warps for given shared memory size.

Instance values map from number of warps (uint64) to value (uint64).

launch__registers_per_thread Number of registers allocated per thread.

launch__registers_per_thread_allocated Number of registers allocated per thread.

launch__shared_mem_config_size Shared memory size configured for the kernel launch. The size depends on the

static, dynamic, and driver shared memory requirements as well as the specified or

platform-determined configuration size.

launch__shared_mem_per_block_allocated Allocated shared memory size per block.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 34
 Metrics Reference

launch__shared_mem_per_block_driver Shared memory size per block, allocated for the CUDA driver.

launch__shared_mem_per_block_dynamic Dynamic shared memory size per block, allocated for the kernel.

launch__shared_mem_per_block_static Static shared memory size per block, allocated for the kernel.

launch__stream_id CUDA stream id for the kernel launch.

launch__thread_count Total number of threads across all blocks for the kernel launch.

launch__uses_cdp Set to 1 if any function object in the launched workload can use CUDA dynamic

parallelism.

launch__waves_per_multiprocessor Number of waves per SM. Partial waves can lead to tail effects where some SMs

become idle while others still have pending work to complete.

NVLink Topology Metrics

nvlink__bandwidth Link bandwidth in bytes/s.

Instance values map from logical nvlink ID (uint64) to value (double).

nvlink__count_logical Total number of logical NVLinks.

nvlink__count_physical Total number of physical links.

Instance values map from physical nvlink device ID (uint64) to value (uint64).

nvlink__destination_ports Destination port numbers (as strings).

Instance values map from logical nvlink ID (uint64) to comma-separated list of port

numbers (string).

nvlink__dev0Id ID of the first connected device.

Instance values map from logical nvlink ID (uint64) to value (uint64).

nvlink__dev0type Type of the first connected device.

Instance values map from logical nvlink ID (uint64) to values [1=GPU, 2=CPU]

(uint64).

nvlink__dev1Id ID of the second connected device.

Instance values map from logical nvlink ID (uint64) to value (uint64).

nvlink__dev1type Type of the second connected device.

Instance values map from logical nvlink ID (uint64) to values [1=GPU, 2=CPU]

(uint64).

nvlink__dev_display_name_all Device display name.

Instance values map from logical nvlink device ID (uint64) to value (string).

nvlink__enabled_mask NVLink enablement mask, per device.

Instance values map from physical nvlink device ID (uint64) to value (uint64).

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 35
 Metrics Reference

nvlink__is_direct_link Indicates, per NVLink, if the link is direct.

Instance values map from logical nvlink ID (uint64) to value (uint64).

nvlink__is_nvswitch_connected Indicates if NVSwitch is connected.

nvlink__max_count Maximum number of NVLinks.

Instance values map from physical nvlink device ID (uint64) to value (uint64).

nvlink__peer_access Indicates if peer access is supported.

Instance values map from logical nvlink ID (uint64) to value (uint64).

nvlink__peer_atomic Indicates if peer atomics are supported.

Instance values map from logical nvlink ID (uint64) to value (uint64).

nvlink__source_ports Source port numbers (as strings).

Instance values map from logical nvlink ID (uint64) to comma-separated list of port

numbers (string).

nvlink__system_access Indicates if system access is supported.

Instance values map from logical nvlink ID (uint64) to value (uint64).

nvlink__system_atomic Indicates if system atomics are supported.

Instance values map from logical nvlink ID (uint64) to value (uint64).

NUMA Topology Metrics

numa__cpu_affinity CPU affinity for each device.

Instance values map from device ID (uint64) to comma-separated values (string).

numa__dev_display_name_all Device display names for all devices.

Instance values map from device ID (uint64) to comma-separated values (string).

numa__id_cpu NUMA ID of the nearest CPU for each device.

Instance values map from device ID (uint64) to comma-separated values (string).

numa__id_memory NUMA ID of the nearest memory for each device.

Instance values map from device ID (uint64) to comma-separated values (string).

Device Attributes
device__attribute_* metrics represent CUDA device attributes. Collecting them
does not require an addition kernel replay pass, as their value is available from the
CUDA driver for each CUDA device.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 36
 Metrics Reference

Warp Stall Reasons

Collected using warp scheduler state sampling. They are incremented regardless if the
scheduler issued an instruction in the same cycle or not. These metrics have instance
values mapping from the function address (uint64) to the number of samples (uint64).

smsp__pcsamp_warps_issue_stalled_barrier Warp was stalled waiting for sibling warps at a CTA barrier. A high number of

warps waiting at a barrier is commonly caused by diverging code paths before a

barrier. This causes some warps to wait a long time until other warps reach the

synchronization point. Whenever possible, try to divide up the work into blocks of

uniform workloads. If the block size is 512 threads or greater, consider splitting it

into smaller groups. This can increase eligible warps without affecting occupancy,

unless shared memory becomes a new occupancy limiter. Also, try to identify which

barrier instruction causes the most stalls, and optimize the code executed before

that synchronization point first.

smsp__pcsamp_warps_issue_stalled_branch_resolving Warp was stalled waiting for a branch target to be computed, and the warp

program counter to be updated. To reduce the number of stalled cycles, consider

using fewer jump/branch operations and reduce control flow divergence, e.g.

by reducing or coalescing conditionals in your code. See also the related No

Instructions state.

smsp__pcsamp_warps_issue_stalled_dispatch_stall Warp was stalled waiting on a dispatch stall. A warp stalled during dispatch has an

instruction ready to issue, but the dispatcher holds back issuing the warp due to

other conflicts or events.

smsp__pcsamp_warps_issue_stalled_drain Warp was stalled after EXIT waiting for all outstanding memory operations to

complete so that warp's resources can be freed. A high number of stalls due to

draining warps typically occurs when a lot of data is written to memory towards the

end of a kernel. Make sure the memory access patterns of these store operations

are optimal for the target architecture and consider parallelized data reduction, if

applicable.

smsp__pcsamp_warps_issue_stalled_imc_miss Warp was stalled waiting for an immediate constant cache (IMC) miss. A read

from constant memory costs one memory read from device memory only on a

cache miss; otherwise, it just costs one read from the constant cache. Immediate

constants are encoded into the SASS instruction as 'c[bank][offset]'. Accesses to

different addresses by threads within a warp are serialized, thus the cost scales

linearly with the number of unique addresses read by all threads within a warp. As

such, the constant cache is best when threads in the same warp access only a few

distinct locations. If all threads of a warp access the same location, then constant

memory can be as fast as a register access.

smsp__pcsamp_warps_issue_stalled_lg_throttle Warp was stalled waiting for the L1 instruction queue for local and global (LG)

memory operations to be not full. Typically, this stall occurs only when executing

local or global memory instructions extremely frequently. Avoid redundant

global memory accesses. Try to avoid using thread-local memory by checking

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 37
 Metrics Reference

if dynamically indexed arrays are declared in local scope, of if the kernel has

excessive register pressure causing by spills. If applicable, consider combining

multiple lower-width memory operations into fewer wider memory operations and

try interleaving memory operations and math instructions.

smsp__pcsamp_warps_issue_stalled_long_scoreboard Warp was stalled waiting for a scoreboard dependency on a L1TEX (local, global,

surface, texture) operation. Find the instruction producing the data being waited

upon to identify the culprit. To reduce the number of cycles waiting on L1TEX data

accesses verify the memory access patterns are optimal for the target architecture,

attempt to increase cache hit rates by increasing data locality (coalescing), or by

changing the cache configuration. Consider moving frequently used data to shared

memory.

smsp__pcsamp_warps_issue_stalled_math_pipe_throttle Warp was stalled waiting for the execution pipe to be available. This stall occurs

when all active warps execute their next instruction on a specific, oversubscribed

math pipeline. Try to increase the number of active warps to hide the existent

latency or try changing the instruction mix to utilize all available pipelines in a

more balanced way.

smsp__pcsamp_warps_issue_stalled_membar Warp was stalled waiting on a memory barrier. Avoid executing any unnecessary

memory barriers and assure that any outstanding memory operations are fully

optimized for the target architecture.

smsp__pcsamp_warps_issue_stalled_mio_throttle Warp was stalled waiting for the MIO (memory input/output) instruction queue

to be not full. This stall reason is high in cases of extreme utilization of the MIO

pipelines, which include special math instructions, dynamic branches, as well as

shared memory instructions. When caused by shared memory accesses, trying to

use fewer but wider loads can reduce pipeline pressure.

smsp__pcsamp_warps_issue_stalled_misc Warp was stalled for a miscellaneous hardware reason.

smsp__pcsamp_warps_issue_stalled_no_instructions Warp was stalled waiting to be selected to fetch an instruction or waiting on an

instruction cache miss. A high number of warps not having an instruction fetched

is typical for very short kernels with less than one full wave of work in the grid.

Excessively jumping across large blocks of assembly code can also lead to more

warps stalled for this reason, if this causes misses in the instruction cache. See also

the related Branch Resolving state.

smsp__pcsamp_warps_issue_stalled_not_selected Warp was stalled waiting for the micro scheduler to select the warp to issue. Not

selected warps are eligible warps that were not picked by the scheduler to issue

that cycle as another warp was selected. A high number of not selected warps

typically means you have sufficient warps to cover warp latencies and you may

consider reducing the number of active warps to possibly increase cache coherence

and data locality.

smsp__pcsamp_warps_issue_stalled_selected Warp was selected by the micro scheduler and issued an instruction.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 38
 Metrics Reference

smsp__pcsamp_warps_issue_stalled_short_scoreboard Warp was stalled waiting for a scoreboard dependency on a MIO (memory input/

output) operation (not to L1TEX). The primary reason for a high number of stalls

due to short scoreboards is typically memory operations to shared memory. Other

reasons include frequent execution of special math instructions (e.g. MUFU) or

dynamic branching (e.g. BRX, JMX). Consult the Memory Workload Analysis section

to verify if there are shared memory operations and reduce bank conflicts, if

reported. Assigning frequently accessed values to variables can assist the compiler

in using low-latency registers instead of direct memory accesses.

smsp__pcsamp_warps_issue_stalled_sleeping Warp was stalled due to all threads in the warp being in the blocked, yielded, or

sleep state. Reduce the number of executed NANOSLEEP instructions, lower the

specified time delay, and attempt to group threads in a way that multiple threads

in a warp sleep at the same time.

smsp__pcsamp_warps_issue_stalled_tex_throttle Warp was stalled waiting for the L1 instruction queue for texture operations to

be not full. This stall reason is high in cases of extreme utilization of the L1TEX

pipeline. Try issuing fewer texture fetches, surface loads, surface stores, or

decoupled math operations. If applicable, consider combining multiple lower-

width memory operations into fewer wider memory operations and try interleaving

memory operations and math instructions. Consider converting texture lookups

or surface loads into global memory lookups. Texture can accept four threads'

requests per cycle, whereas global accepts 32 threads.

smsp__pcsamp_warps_issue_stalled_wait Warp was stalled waiting on a fixed latency execution dependency. Typically, this

stall reason should be very low and only shows up as a top contributor in already

highly optimized kernels. Try to hide the corresponding instruction latencies by

increasing the number of active warps, restructuring the code or unrolling loops.

Furthermore, consider switching to lower-latency instructions, e.g. by making use

of fast math compiler options.

Warp Stall Reasons (Not Issued)

Collected using warp scheduler state sampling. Thye are incremented only on cycles
in which the warp scheduler issued no instruction. These metrics have instance values
mapping from the function address (uint64) to the number of samples (uint64).

smsp__pcsamp_warps_issue_stalled_barrier_not_issued Warp was stalled waiting for sibling warps at a CTA barrier. A high number of

warps waiting at a barrier is commonly caused by diverging code paths before a

barrier. This causes some warps to wait a long time until other warps reach the

synchronization point. Whenever possible, try to divide up the work into blocks of

uniform workloads. If the block size is 512 threads or greater, consider splitting it

into smaller groups. This can increase eligible warps without affecting occupancy,

unless shared memory becomes a new occupancy limiter. Also, try to identify which

barrier instruction causes the most stalls, and optimize the code executed before

that synchronization point first.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 39
 Metrics Reference

smsp__pcsamp_warps_issue_stalled_branch_resolving_not_issued Warp was stalled waiting for a branch target to be computed, and the warp

program counter to be updated. To reduce the number of stalled cycles, consider

using fewer jump/branch operations and reduce control flow divergence, e.g.

by reducing or coalescing conditionals in your code. See also the related No

Instructions state.

smsp__pcsamp_warps_issue_stalled_dispatch_stall_not_issued Warp was stalled waiting on a dispatch stall. A warp stalled during dispatch has an

instruction ready to issue, but the dispatcher holds back issuing the warp due to

other conflicts or events.

smsp__pcsamp_warps_issue_stalled_drain_not_issued Warp was stalled after EXIT waiting for all memory operations to complete so

that warp resources can be freed. A high number of stalls due to draining warps

typically occurs when a lot of data is written to memory towards the end of a

kernel. Make sure the memory access patterns of these store operations are

optimal for the target architecture and consider parallelized data reduction, if

applicable.

smsp__pcsamp_warps_issue_stalled_imc_miss_not_issued Warp was stalled waiting for an immediate constant cache (IMC) miss. A read

from constant memory costs one memory read from device memory only on a

cache miss; otherwise, it just costs one read from the constant cache. Accesses to

different addresses by threads within a warp are serialized, thus the cost scales

linearly with the number of unique addresses read by all threads within a warp. As

such, the constant cache is best when threads in the same warp access only a few

distinct locations. If all threads of a warp access the same location, then constant

memory can be as fast as a register access.

smsp__pcsamp_warps_issue_stalled_lg_throttle_not_issued Warp was stalled waiting for the L1 instruction queue for local and global (LG)

memory operations to be not full. Typically, this stall occurs only when executing

local or global memory instructions extremely frequently. Avoid redundant

global memory accesses. Try to avoid using thread-local memory by checking

if dynamically indexed arrays are declared in local scope, of if the kernel has

excessive register pressure causing by spills. If applicable, consider combining

multiple lower-width memory operations into fewer wider memory operations and

try interleaving memory operations and math instructions.

smsp__pcsamp_warps_issue_stalled_long_scoreboard_not_issued Warp was stalled waiting for a scoreboard dependency on a L1TEX (local, global,

surface, texture) operation. Find the instruction producing the data being waited

upon to identify the culprit. To reduce the number of cycles waiting on L1TEX data

accesses verify the memory access patterns are optimal for the target architecture,

attempt to increase cache hit rates by increasing data locality (coalescing), or by

changing the cache configuration. Consider moving frequently used data to shared

memory.

smsp__pcsamp_warps_issue_stalled_math_pipe_throttle_not_issued Warp was stalled waiting for the execution pipe to be available. This stall occurs

when all active warps execute their next instruction on a specific, oversubscribed

math pipeline. Try to increase the number of active warps to hide the existent

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 40
 Metrics Reference

latency or try changing the instruction mix to utilize all available pipelines in a

more balanced way.

smsp__pcsamp_warps_issue_stalled_membar_not_issued Warp was stalled waiting on a memory barrier. Avoid executing any unnecessary

memory barriers and assure that any outstanding memory operations are fully

optimized for the target architecture.

smsp__pcsamp_warps_issue_stalled_mio_throttle_not_issued Warp was stalled waiting for the MIO (memory input/output) instruction queue

to be not full. This stall reason is high in cases of extreme utilization of the MIO

pipelines, which include special math instructions, dynamic branches, as well as

shared memory instructions. When caused by shared memory accesses, trying to

use fewer but wider loads can reduce pipeline pressure.

smsp__pcsamp_warps_issue_stalled_misc_not_issued Warp was stalled for a miscellaneous hardware reason.

smsp__pcsamp_warps_issue_stalled_no_instructions_not_issued Warp was stalled waiting to be selected to fetch an instruction or waiting on an

instruction cache miss. A high number of warps not having an instruction fetched

is typical for very short kernels with less than one full wave of work in the grid.

Excessively jumping across large blocks of assembly code can also lead to more

warps stalled for this reason, if this causes misses in the instruction cache. See also

the related Branch Resolving state.

smsp__pcsamp_warps_issue_stalled_not_selected_not_issued Warp was stalled waiting for the micro scheduler to select the warp to issue. Not

selected warps are eligible warps that were not picked by the scheduler to issue

that cycle as another warp was selected. A high number of not selected warps

typically means you have sufficient warps to cover warp latencies and you may

consider reducing the number of active warps to possibly increase cache coherence

and data locality.

smsp__pcsamp_warps_issue_stalled_selected_not_issued Warp was selected by the micro scheduler and issued an instruction.

smsp__pcsamp_warps_issue_stalled_short_scoreboard_not_issued Warp was stalled waiting for a scoreboard dependency on a MIO (memory input/

output) operation (not to L1TEX). The primary reason for a high number of stalls

due to short scoreboards is typically memory operations to shared memory. Other

reasons include frequent execution of special math instructions (e.g. MUFU) or

dynamic branching (e.g. BRX, JMX). Consult the Memory Workload Analysis section

to verify if there are shared memory operations and reduce bank conflicts, if

reported. Assigning frequently accessed values to variables can assist the compiler

in using low-latency registers instead of direct memory accesses.

smsp__pcsamp_warps_issue_stalled_sleeping_not_issued Warp was stalled due to all threads in the warp being in the blocked, yielded, or

sleep state. Reduce the number of executed NANOSLEEP instructions, lower the

specified time delay, and attempt to group threads in a way that multiple threads

in a warp sleep at the same time.

smsp__pcsamp_warps_issue_stalled_tex_throttle_not_issued Warp was stalled waiting for the L1 instruction queue for texture operations to

be not full. This stall reason is high in cases of extreme utilization of the L1TEX

pipeline. Try issuing fewer texture fetches, surface loads, surface stores, or

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 41
 Metrics Reference

decoupled math operations. If applicable, consider combining multiple lower-

width memory operations into fewer wider memory operations and try interleaving

memory operations and math instructions. Consider converting texture lookups

or surface loads into global memory lookups. Texture can accept four threads'

requests per cycle, whereas global accepts 32 threads.

smsp__pcsamp_warps_issue_stalled_wait_not_issued Warp was stalled waiting on a fixed latency execution dependency. Typically, this

stall reason should be very low and only shows up as a top contributor in already

highly optimized kernels. Try to hide the corresponding instruction latencies by

increasing the number of active warps, restructuring the code or unrolling loops.

Furthermore, consider switching to lower-latency instructions, e.g. by making use

of fast math compiler options.

Source Metrics
Collected using SASS-patching. These metrics have instance values mapping
from function address (uint64) to associated values (uint64). Metrics
memory_[access_]type map to string values.

branch_inst_executed Number of unique branch targets assigned to the instruction, including both

divergent and uniform branches.

derived__avg_thread_executed Average number of thread-level executed instructions per warp (regardless of their

predicate). Computed as: thread_inst_executed / inst_executed

derived__avg_thread_executed_true Average number of predicated-on thread-level executed instructions per warp.

Computed as: thread_inst_executed_true / inst_executed

derived__memory_l1_conflicts_shared_nway Average N-way conflict in L1 per shared memory instruction. A 1-way

access has no conflicts and resolves in a single pass. Computed as:

memory_l1_wavefronts_shared / inst_executed

derived__memory_l1_wavefronts_shared_excessive Excessive number of wavefronts in L1 from shared memory instructions, because

not all not predicated-off threads performed the operation.

derived__memory_l2_theoretical_sectors_global_excessive Excessive theoretical number of sectors requested in L2 from global memory

instructions, because not all not predicated-off threads performed the operation.

inst_executed Number of warp-level executed instructions, ignoring instruction predicates.

Warp-level means the values increased by one per individual warp executing the

instruction, independent of the number of participating threads within each warp.

memory_access_size_type The size of the memory access, in bits.

memory_access_type The type of memory access (e.g. load or store).

memory_l1_tag_requests_global Number of L1 tag requests generated by global memory instructions.

memory_l1_wavefronts_shared Number of wavefronts in L1 from shared memory instructions.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 42
 Metrics Reference

memory_l1_wavefronts_shared_ideal Ideal number of wavefronts in L1 from shared memory instructions, assuming each

not predicated-off thread performed the operation.

memory_l2_theoretical_sectors_global Theoretical number of sectors requested in L2 from global memory instructions.

memory_l2_theoretical_sectors_global_ideal Ideal number of sectors requested in L2 from global memory instructions, assuming

each not predicated-off thread performed the operation.

memory_l2_theoretical_sectors_local Theoretical number of sectors requested in L2 from local memory instructions.

memory_type The accessed address space (global/local/shared).

smsp__branch_targets_threads_divergent Number of divergent branch targets, including fallthrough. Incremented only when

there are two or more active threads with divergent targets.

smsp__branch_targets_threads_uniform Number of uniform branch execution, including fallthrough, where all active

threads selected the same branch target.

smsp__pcsamp_sample_count Number of collected samples per program counter from the periodic sampler.

thread_inst_executed Number of thread-level executed instructions, regardless of predicate presence or

evaluation.

thread_inst_executed_true Number of thread-level executed instructions, where the instruction predicate

evaluated to true, or no predicate was given.

L2 Cache Eviction Metrics

smsp__sass_inst_executed_memdesc_explicit_evict_type L2 cache eviction policy types.

smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_first Number of warp-level executed instructions with L2 cache eviction hit property

'first'.

smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_last Number of warp-level executed instructions with L2 cache eviction hit property

'last'.

smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_normal Number of warp-level executed instructions with L2 cache eviction hit property

'normal'.

smsp__sass_inst_executed_memdesc_explicit_hitprop_evict_normal_demote
Number of warp-level executed instructions with L2 cache eviction hit property

'normal demote'.

smsp__sass_inst_executed_memdesc_explicit_missprop_evict_first Number of warp-level executed instructions with L2 cache eviction miss property

'first'.

smsp__sass_inst_executed_memdesc_explicit_missprop_evict_normal Number of warp-level executed instructions with L2 cache eviction miss property

'normal'.

Instructions Per Opcode Metrics

Collected using SASS-patching. These metrics have instance values mapping from the
SASS opcode (string) to the number of executions (uint64).

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 43
 Metrics Reference

sass__inst_executed_per_opcode Number of warp-level executed instructions, instanced by basic SASS opcode.

sass__inst_executed_per_opcode_with_modifier_all Number of warp-level executed instructions, instanced by all SASS opcode

modifiers.

sass__inst_executed_per_opcode_with_modifier_selective Number of warp-level executed instructions, instanced by selective SASS opcode

modifiers.

sass__thread_inst_executed_true_per_opcode Number of thread-level executed instructions, instanced by basic SASS opcode.

sass__thread_inst_executed_true_per_opcode_with_modifier_all Number of thread-level executed instructions, instanced by all SASS opcode

modifiers.

sass__thread_inst_executed_true_per_opcode_with_modifier_selectiveNumber of thread-level executed instructions, instanced by selective SASS opcode

modifiers.

Metric Groups
group:memory__chart Group of metrics for the workload analysis chart.

group:memory__dram_table Group of metrics for the device memory workload analysis table.

group:memory__first_level_cache_table Group of metrics for the L1/TEX cache workload analysis table.

group:memory__l2_cache_evict_policy_table Group of metrics for the L2 cache eviction policies table.

group:memory__l2_cache_table Group of metrics for the L2 cache workload analysis table.

group:memory__shared_table Group of metrics for the shared memory workload analysis table.

group:smsp__pcsamp_warp_stall_reasons Group of metrics for the number of samples from the statistical sampler per

program location.

group:smsp__pcsamp_warp_stall_reasons_not_issued Group of metrics for the number of samples from the statistical sampler per

program location on cycles the warp scheduler issued no instructions.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 44
Chapter 5.
SAMPLING

NVIDIA Nsight Compute supports periodic sampling of the warp program counter and
warp scheduler state on desktop devices of compute capability 6.1 and above.
At a fixed interval of cycles, the sampler in each streaming multiprocessor selects an
active warp and outputs the program counter and the warp scheduler state. The tool
selects the minimum interval for the device. On small devices, this can be every 32
cycles. On larger chips with more multiprocessors, this may be 2048 cycles. The sampler
selects a random active warp. On the same cycle the scheduler may select a different
warp to issue.

5.1. Warp Scheduler States

See the Warp Stall Reasons tables in the Metrics Reference for a description of the
individual warp scheduler states.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 45
Chapter 6.
REPRODUCIBILITY

In order to provide actionable and deterministic results across application runs, NVIDIA
Nsight Compute applies various methods to adjust how metrics are collected. This
includes serializing kernel launches, purging GPU caches before each kernel replay or
adjusting GPU clocks.

6.1. Serialization
NVIDIA Nsight Compute serializes kernel launches within the profiled application,
potentially across multiple processes profiled by one or more instances of the tool at the
same time.
Serialization across processes is necessary since for the collection of HW performance
metrics, some GPU and driver objects can only be acquired by a single process at a time.
To achieve this, the lock file TMPDIR/nsight-compute-lock is used. On Windows,
TMPDIR is the path returned by the Windows GetTempPath API function. On other
platforms, it is the path supplied by the first environment variable in the list TMPDIR,
TMP, TEMP, TEMPDIR. If none of these is found, it's /var/nvidia on QNX and /tmp
otherwise.
Serialization within the process is required for most metrics to be mapped to the proper
kernel. In addition, without serialization, performance metric values might vary widely
if kernel execute concurrently on the same device.
It is currently not possible to disable this tool behavior. Refer to the FAQ entry on
possible workarounds.

6.2. Clock Control

For many metrics, their value is directly influenced by the current GPU SM and memory
clock frequencies. For example, if a kernel instance is profiled that has prior kernel
executions in the application, the GPU might already be in a higher clocked state and
the measured kernel duration, along with other metrics, will be affected. Likewise, if
a kernel instance is the first kernel to be launched in the application, GPU clocks will

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 46
 Reproducibility

regularly be lower. In addition, due to kernel replay, the metric value might depend on
which replay pass it is collected in, as later passes would result in higher clock states.
To mitigate this non-determinism, NVIDIA Nsight Compute attempts to limit GPU
clock frequencies to their base value. As a result, metric values are less impacted by the
location of the kernel in the application, or by the number of the specific replay pass.
However, this behavior might be undesirable for analysis of the kernel, e.g. in cases
where an external tool is used to fix clock frequencies, or where the behavior of the
kernel within the application is analyzed. To solve this, users can adjust the --clock-
control option to specify if any clock frequencies should be fixed by the tool.
Factors affecting Clock Control:
‣ Note that thermal throttling directed by the driver cannot be controlled by the tool
and always overrides any selected options.
‣ On mobile targets, e.g. L4T or QNX, there may be variations in profiling results due
the inability for the tool to lock clocks. Using Nsight Compute’s --clock-control
to set the GPU clocks will fail or will be silently ignored when profiling on a GPU
partition.
‣ On L4T, you can use the jetson_clocks script to lock the clocks at their
maximums during profiling.
‣ See the Special Configurations section for MIG and vGPU clock control.

6.3. Cache Control

As explained in Kernel Replay, the kernel might need to be replayed multiple times to
collect all requested metrics. While NVIDIA Nsight Compute can save and restore the
contents of GPU device memory accessed by the kernel for each pass, it cannot do the
same for the contents of HW caches, such as e.g. the L1 and L2 cache.
This can have the effect that later replay passes might have better or worse performance
than e.g. the first pass, as the caches could already be primed with the data last accessed
by the kernel. Similarly, the values of HW performance counters collected by the first
pass might depend on which kernels, if any, were executed prior to the measured kernel
launch.
In order to make HW performance counter value more deterministic, NVIDIA Nsight
Compute by default flushes all GPU caches before each replay pass. As a result, in each
pass, the kernel will access a clean cache and the behavior will be as if the kernel was
executed in complete isolation.
This behavior might be undesirable for performance analysis, especially if the
measurement focuses on a kernel within a larger application execution, and if the
collected data targets cache-centric metrics. In this case, you can use --cache-control
none to disable flushing of any HW cache by the tool.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 47
 Reproducibility

6.4. Persistence Mode

The NVIDIA kernel mode driver must be running and connected to a target GPU device
before any user interactions with that device can take place. The driver behavior differs
depending on the OS. Generally, on Linux, if the kernel mode driver is not already
running or connected to a target GPU, the invocation of any program that attempts to
interact with that GPU will transparently cause the driver to load and/or initialize the
GPU. When all GPU clients terminate the driver will then deinitialize the GPU.
If persistence mode is not enabled (as part of the OS, or by the user), applications
triggering GPU initialization may incur a short startup cost. In addition, on some
configurations, there may also be a shutdown cost when the GPU is de-initialized at the
end of the application.
It is recommended to enable persistence mode on applicable operating systems before
profiling with NVIDIA Nsight Compute for more consistent application behavior.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 48
Chapter 7.
SPECIAL CONFIGURATIONS

7.1. Multi Instance GPU

Multi-Instance GPU (MIG) is a feature that allows a GPU to be partitioned into multiple
CUDA devices. The partitioning is carried out on two levels: First, a GPU can be split
into one or multiple GPU Instances. Each GPU Instance claims ownership of one or
more streaming multiprocessors (SM), a subset of the overall GPU memory, and possibly
other GPU resources, such as the video encoders/decoders. Second, each GPU Instance
can be further partitioned into one or more Compute Instances. Each Compute Instance
has exclusive ownership of its assigned SMs of the GPU Instance. However, all Compute
Instances within a GPU Instance share the GPU Instance's memory and memory
bandwidth. Every Compute Instance acts and operates as a CUDA device with a unique
device ID. See the driver release notes as well as the documentation for the nvidia-smi
CLI tool for more information on how to configure MIG instances.
For profiling, a Compute Instance can be of one of two types: isolated or shared.
An isolated Compute Instance owns all of its assigned resources and does not share any
GPU unit with another Compute Instance. In other words, the Compute Instance is the
same size as its parent GPU Instance and consequently does not have any other sibling
Compute Instances. Profiling works as usual for isolated Compute Instances.
A shared Compute Instance uses GPU resources that can potentially also be accessed
by other Compute Instances in the same GPU Instance. Due to this resource sharing,
collecting profiling data from those shared units is not permitted. Attempts to collect
metrics from a shared unit fail with an error message of ==ERROR== Failed to
access the following metrics. When profiling on a MIG instance, it
is not possible to collect metrics from GPU units that are shared
with other MIG instances followed by the list of failing metrics. Collecting only
metrics from GPU units that are exclusively owned by a shared Compute Instance is still
possible.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 49
 Special Configurations

Locking Clocks
NVIDIA Nsight Compute is not able to set the clock frequency on any Compute Instance
for profiling. You can continue analyzing kernels without fixed clock frequencies (using
--clock-control none; see here for more details). If you have sufficient permissions,
nvidia-smi can be used to configure a fixed frequency for the whole GPU by calling
nvidia-smi --lock-gpu-clocks=tdp,tdp. This sets the GPU clocks to the base TDP
frequency until you reset the clocks by calling nvidia-smi --reset-gpu-clocks.

MIG on Baremetal (non-vGPU)

All Compute Instances on a GPU share the same clock frequencies.

MIG on NVIDIA vGPU

Enabling profiling for a VM gives the VM access to the GPU's global performance
counters, which may include activity from other VMs executing on the same GPU.
Enabling profiling for a VM also allows the VM to lock clocks on the GPU, which
impacts all other VMs executing on the same GPU, including MIG Compute Instances.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 50
Chapter 8.
ROOFLINE CHARTS

Roofline charts provide a very helpful way to visualize achieved performance on

complex processing units, like GPUs. This section introduces the Roofline charts that are
presented within a profile report.

8.1. Overview
Kernel performance is not only dependent on the operational speed of the GPU. Since a
kernel requires data to work on, performance is also dependent on the rate at which the
GPU can feed data to the kernel. A typical roofline chart combines the peak performance
and memory bandwidth of the GPU, with a metric called Arithmetic Intensity (a ratio
between Work and Memory Traffic), into a single chart, to more realistically represent the
achieved performance of the profiled kernel. A simple roofline chart might look like the
following:

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 51
 Roofline Charts

This chart actually shows two different rooflines. However, the following components
can be identified for each:
‣ Vertical Axis - The vertical axis represents Floating Point Operations per Second
(FLOPS). For GPUs this number can get quite large and so the numbers on this axis
can be scaled for easier reading (as shown here). In order to better accommodate the
range, this axis is rendered using a logarithmic scale.
‣ Horizontal Axis - The horizontal axis represents Arithmetic Intensity, which is
the ratio between Work (expressed in floating point operations per second), and
Memory Traffic (expressed in bytes per second). The resulting unit is in floating point
operations per byte. This axis is also shown using a logarithmic scale.
‣ Memory Bandwidth Boundary - The memory bandwidth boundary is the sloped part
of the roofline. By default, this slope is determined entirely by the memory transfer
rate of the GPU but can be customized inside the SpeedOfLight_RooflineChart.section
file if desired.
‣ Peak Performance Boundary - The peak performance boundary is the flat part of
the roofline By default, this value is determined entirely by the peak performance of
the GPU but can be customized inside the SpeedOfLight_RooflineChart.section file if
desired.
‣ Ridge Point - The ridge point is the point at which the memory bandwidth
boundary meets the peak performance boundary. This point is a useful reference
when analyzing kernel performance.
‣ Achieved Value - The achieved value represents the performance of the profiled
kernel. If baselines are being used, the roofline chart will also contain an achieved
value for each baseline. The outline color of the plotted achieved value point can be
used to determine from which baseline the point came.

8.2. Analysis
The roofline chart can be very helpful in guiding performance optimization efforts for a
particular kernel.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 52
 Roofline Charts

As shown here, the ridge point partitions the roofline chart into two regions. The area
shaded in blue under the sloped Memory Bandwidth Boundary is the Memory Bound
region, while the area shaded in green under the Peak Performance Boundary is the
Compute Bound region. The region in which the achieved value falls, determines the
current limiting factor of kernel performance.
The distance from the achieved value to the respective roofline boundary (shown
in this figure as a dotted white line), represents the opportunity for performance
improvement. The closer the achieved value is to the roofline boundary, the more optimal
is its performance. An achieved value that lies on the Memory Bandwidth Boundary but is
not yet at the height of the ridge point would indicate that any further improvements in
overall FLOP/s are only possible if the Arithmetic Intensity is increased at the same time.
Using the baseline feature in combination with roofline charts, is a good way to track
optimization progress over a number of kernel executions.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 53
Chapter 9.
MEMORY CHART

The Memory Chart shows a graphical, logical representation of performance data for
memory subunits on and off the GPU. Performance data includes transfer sizes, hit rates,
number of instructions or requests, etc.

9.1. Overview

Logical Units (green)

Logical units are shown in green color.

‣ Kernel: The CUDA kernel executing on the GPU's Streaming Multiprocessors

‣ Global: CUDA global memory
‣ Local: CUDA local memory
‣ Texture: CUDA texture memory
‣ Surface: CUDA surface memory
‣ Shared: CUDA shared memory

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 54
 Memory Chart

‣ Load Global Store Shared: Instructions loading directly from global into shared
memory without intermediate register file access

Physical Units (blue)

Physical units are shown in blue color.

‣ L1/TEX Cache: The L1/Texture cache. The underlying physical memory is split
between this cache and the user-managed Shared Memory.
‣ Shared Memory: CUDA's user-managed shared memory. The underlying physical
memory is split between this and the L1/TEX Cache.
‣ L2 Cache: The L2 cache
‣ L2 Compression: The memory compression unit of the L2 Cache
‣ System Memory: Off-chip system (CPU) memory
‣ Device Memory: On-chip device (GPU) memory of the CUDA device that executes
the kernel
‣ Peer Memory: On-chip device (GPU) memory of other CUDA devices
Depending on the exact GPU architecture, the exact set of shown units can vary, as not
all GPUs have all units.

Links
Links between Kernel and other logical units represent the number of executed
instructions (Inst) targeting the respective unit. For example, the link between Kernel and
Global represents the instructions loading from or storing to the global memory space.
Instructions using the NVIDIA A100's Load Global Store Shared paradigm are shown
separately, as their register or cache access behavior can be different from regular global
loads or shared stores.
Links between logical units and blue, physical units represent the number of requests
(Req) issued as a result of their respective instructions. For example, the link going from
L1/TEX Cache to Global shows the number of requests generated due to global load
instructions.
The color of each link represents the percentage of peak utilization of the corresponding
communication path. The color legend to the right of the chart shows the applied color
gradient from unused (0%) to operating at peak performance (100%). Triangle markers
to the left of the legend correspond to the links in the chart. The markers offer a more
accurate value estimate for the achieved peak performances than the color gradient
alone.
A unit often shares a common data port for incoming and outgoing traffic. While the
links sharing a port might operate well below their individual peak performances, the
unit's data port may have already reached its peak. Port utilization is shown in the chart
by colored rectangles inside the units located at the incoming and outgoing links. Ports

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 55
 Memory Chart

use the same color gradient as the data links and have also a corresponding marker to
the left of the legend.
An example of the correlation between the peak values reported in the memory tables
and the ports in the memory chart is shown below.

Metrics
Metrics from this chart can be collected on the command line using --set full, --
section MemoryWorkloadAnalysis_Chart or --metrics group:memory__chart.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 56
Chapter 10.
MEMORY TABLES

The Memory Tables show detailed metrics for the various memory HW units, such as
shared memory, the caches, and device memory. For most table entries, you can hover
over it to see the underlying metric name and description. Some entries are generated
as derivatives from other cells, and do not show a metric name on their own, but the
respective calculation. If a certain metric does not contribute to the generic derivative
calculation, it is shown as UNUSED in the tooltip. You can hover over row or column
headers to see a description of this part of the table.

10.1. Shared Memory

Columns

Instructions For each access type, the total number

of all actually executed assembly (SASS)
instructions per warp. Predicated-off
instructions are not included.
E.g., the instruction STS would be counted
towards Shared Store.
Requests The total number of all requests to shared
memory. On SM 7.0 (Volta) and newer

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 57
 Memory Tables

architectures, each shared memory

instruction generates exactly one request.
Wavefronts Number of wavefronts required to service
the requested shared memory data.
Wavefronts are serialized and processed
on different cycles.
% Peak Percentage of peak utilization. Higher
values imply a higher utilization of the
unit and can show potential bottlenecks,
as it does not necessarily indicate efficient
usage.
Bank Conflicts If multiple threads' requested addresses
map to different offsets in the same
memory bank, the accesses are serialized.
The hardware splits a conflicting memory
request into as many separate conflict-
free requests as necessary, decreasing the
effective bandwidth by a factor equal to
the number of colliding memory requests.

Rows

(Access Types) Shared memory access operations.

Total The aggregate for all access types in the
same column.

Metrics
Metrics from this table can be collected on the command line using --set
full, --section MemoryWorkloadAnalysis_Tables or --metrics
group:memory__shared_table.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 58
 Memory Tables

10.2. L1/TEX Cache

Columns

Instructions For each access type, the total number

of all actually executed assembly (SASS)
instructions per warp. Predicated-off
instructions are not included.
E.g., the instruction LDG would be
counted towards Global Loads.
Requests The total number of all requests to L1,
generated for each instruction type. On
SM 7.0 (Volta) and newer architectures,
each instruction generates exactly one
request for LSU traffic (global, local, ...).

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 59
 Memory Tables

For texture (TEX) traffic, more than one

request may be generated.
In the example, each of the 65536 global
load instructions generates exactly one
request.
Wavefronts Number of wavefronts required to
service the requested memory operation.
Wavefronts are serialized and processed
on different cycles.
Wavefront % Peak Percentage of peak utilization for the units
processing wavefronts. High numbers can
imply that the processing pipelines are
saturated and can become a bottleneck.
Sectors The total number of all L1 sectors accesses
sent to L1. Each load or store request
accesses one or more sectors in the L1
cache. Atomics and reductions are passed
through to the L2 cache.
Sectors/Req The average ratio of sectors to requests
for the L1 cache. For the same number of
active threads in a warp, smaller numbers
imply a more efficient memory access
pattern. For warps with 32 active threads,
the optimal ratios per access size are: 32-
bit: 4, 64-bit: 8, 128-bit: 16. Smaller ratios
indicate some degree of uniformity or
overlapped loads within a cache line.
Higher numbers can imply uncoalesced
memory accesses and will result in
increased memory traffic.
In the example, the average ratio for global
loads is 32 sectors per request, which
implies that each thread needs to access a
different sector. Ideally, for warps with 32
active threads, with each thread accessing
a single, aligned 32-bit value, the ratio
would be 4, as every 8 consecutive threads
access the same sector.
Hit Rate Sector hit rate (percentage of requested
sectors that do not miss) in the L1 cache.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 60
 Memory Tables

Sectors that miss need to be requested

from L2, thereby contributing to Sector
Misses to L2. Higher hit rates imply better
performance due to lower access latencies,
as the request can be served by L1 instead
of a later stage. Not to be confused with
Tag Hit Rate (not shown).
Bytes Total number of bytes requested from L1.
This is identical to the number of sectors
multiplied by 32 byte, since the minimum
access size in L1 is one sector.
Sector Misses to L2 Total number of sectors that miss in L1
and generate subsequent requests in the
L2 Cache.
In this example, the 262144 sector misses
for global and local loads can be computed
as the miss-rate of 12.5%, multiplied by the
number of 2097152 sectors.
% Peak to L2 Percentage of peak utilization of the
L1-to-XBAR interface, used to send L2
cache requests. If this number is high, the
workload is likely dominated by scattered
{writes, atomics, reductions}, which can
increase the latency and cause warp stalls.
Returns to SM Number of return packets sent from the
L1 cache back to the SM. Larger request
access sizes result in higher number of
returned packets.
% Peak to SM Percentage of peak utilization of the
XBAR-to-L1 return path (compare
Returns to SM). If this number is high,
the workload is likely dominated by
scattered reads, thereby causing warp
stalls. Improving read-coalescing or the L1
hit rate could reduce this utilization.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 61
 Memory Tables

Rows

(Access Types) The various access types, e.g. loads from

global memory or reduction operations on
surface memory.
Loads The aggregate of all load access types in
the same column.
Stores The aggregate of all store access types in
the same column.
Total The aggregate of all load and store access
types in the same column.

Metrics
Metrics from this table can be collected on the command line using --set
full, --section MemoryWorkloadAnalysis_Tables or --metrics
group:memory__first_level_cache_table.

10.3. L2 Cache

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 62
 Memory Tables

Columns

Requests For each access type, the total number

of requests made to the L2 cache. This
correlates with the Sector Misses to L2 for
the L1 cache. Each request targets one 128
byte cache line.
Sectors For each access type, the total number of
sectors requested from the L2 cache. Each
request accesses one or more sectors.
Sectors/Req The average ratio of sectors to requests
for the L2 cache. For the same number of
active threads in a warp, smaller numbers
imply a more efficient memory access
pattern. For warps with 32 active threads,
the optimal ratios per access size are: 32-
bit: 4, 64-bit: 8, 128-bit: 16. Smaller ratios
indicate some degree of uniformity or
overlapped loads within a cache line.
Higher numbers can imply uncoalesced
memory accesses and will result in
increased memory traffic.
% Peak Percentage of peak sustained number
of sectors. The "work package" in the L2
cache is a sector. Higher values imply
a higher utilization of the unit and can
show potential bottlenecks, as it does not
necessarily indicate efficient usage.
Hit Rate Hit rate (percentage of requested sectors
that do not miss) in the L2 cache. Sectors
that miss need to be requested from a later
stage, thereby contributing to one of Sector
Misses to Device, Sector Misses to System, or
Sector Misses to Peer. Higher hit rates imply
better performance due to lower access
latencies, as the request can be served by
L2 instead of a later stage.
Bytes Total number of bytes requested from L2.
This is identical to the number of sectors
multiplied by 32 byte, since the minimum
access size in L2 is one sector.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 63
 Memory Tables

Throughput Achieved L2 cache throughput in bytes

per second. High values indicate high
utilization of the unit.
Sector Misses to Device Total number of sectors that miss in L2
and generate subsequent requests in
device memory.
Sector Misses to System Total number of sectors that miss in L2
and generate subsequent requests in
system memory.
Sector Misses to Peer Total number of sectors that miss in L2
and generate subsequent requests in peer
memory.

Rows

(Access Types) The various access types, e.g. loads or

reductions originating from L1 cache.
L1/TEX Total Total for all operations originating from
the L1 cache.
ECC Total Total for all operations caused by ECC
(Error Correction Code). If ECC is enabled,
L2 write requests that partially modify
a sector cause a corresponding sector
load from DRAM. These additional load
operations increase the sector misses of L2.
L2 Fabric Total Total for all operations across the L2 fabric
connecting the two L2 partitions. This
row is only shown for kernel launches on
CUDA devices with L2 fabric.
GPU Total Total for all operations across all clients of
the L2 cache. Independent of having them
split out separately in this table.

Metrics
Metrics from this table can be collected on the command line using --set
full, --section MemoryWorkloadAnalysis_Tables or --metrics
group:memory__l2_cache_table.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 64
 Memory Tables

10.4. L2 Cache Eviction Policies

Columns

First Number of sectors accessed in the L2

cache using the evict_first policy. Data
cached with this policy will be first in the
eviction priority order and will likely be
evicted when cache eviction is required.
This policy is suitable for streaming data.
Hit Rate Cache hit rate for sector accesses in the L2
cache using the evict_first policy.
Last Number of sectors accessed in the L2
cache using the evict_last policy. Data
cached with this policy will be last in the
eviction priority order and will likely
be evicted only after other data with
evict_normal or evict_first eviction
policy is already evicted. This policy
is suitable for data that should remain
persistent in cache.
Hit Rate Cache hit rate for sector accesses in the L2
cache using the evict_last policy.
Normal Number of sectors accessed in the L2
cache using the evict_normal policy.
This is the default policy.
Hit Rate Cache hit rate for sector accesses in the L2
cache using the evict_normal policy.
Normal Demote Number of sectors accessed in the L2
cache using the evict_normal_demote
policy.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 65
 Memory Tables

Hit Rate Cache hit rate for sector accesses in the L2

cache using the evict_normal_demote
policy.

Rows

(Access Types) The various access types, e.g. loads or

reductions, originating from L1 cache.
L1/TEX Total Total for all operations originating from
the L1 cache.
L2 Fabric Total Total for all operations across the L2 fabric
connecting the two L2 partitions. This
row is only shown for kernel launches on
CUDA devices with L2 fabric.
GPU Total Total for all operations across all clients of
the L2 cache. Independent of having them
split out separately in this table.

Metrics
Metrics from this table can be collected on the command line using --set
full, --section MemoryWorkloadAnalysis_Tables or --metrics
group:memory__l2_cache_evict_policy_table. Note that this table is only
available on GPUs with GA100 or newer.

10.5. Device Memory

Columns

Sectors For each access type, the total number of

sectors requested from device memory.
% Peak Percentage of peak device memory
utilization. Higher values imply a higher
utilization of the unit and can show
potential bottlenecks, as it does not
necessarily indicate efficient usage.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 66
 Memory Tables

Bytes Total number of bytes transferred between

L2 Cache and device memory.
Throughput Achieved device memory throughput in
bytes per second. High values indicate
high utilization of the unit.

Rows

(Access Types) Device memory loads and stores.

Total The aggregate for all access types in the
same column.

Metrics
Metrics from this table can be collected on the command line using --set
full, --section MemoryWorkloadAnalysis_Tables or --metrics
group:memory__dram_table.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 67
Chapter 11.
FAQ

‣ n/a metric values

n/a means that the metric value is "not available". The most common reason is
that the requested metric does not exist. This can either be the result of a typo, or
a missing suffix. Verify the metric name against the output of of the --query-
metrics NVIDIA Nsight Compute CLI option.
If the metric name was copied (e.g. from an old version of this documentation),
make sure that it does not contain zero-width unicode characters.
Finally, the metric might simply not exist for the targeted
GPU architecture. For example, the IMMA pipeline metric
sm__inst_executed_pipe_tensor_op_imma.avg.pct_of_peak_sustained_active
is not available on GV100 chips.
‣ Metric values outside the expected logical range
This includes e.g. percentages exceeding 100% or metrics reporting negative values.
For further details, see Range and Precision.
‣ ERR_NVGPUCTRPERM - The user does not have permission to access NVIDIA
GPU Performance Counters on the target device.
By default, NVIDIA drivers require elevated permissions to access GPU
performance counters. On mobile platforms, profile as root/using sudo. On
other platforms, you can either start profiling as root/using sudo, or by enabling
non-admin profiling. For further details, see https://developer.nvidia.com/
ERR_NVGPUCTRPERM.
On Windows Subsystem for Linux (WSL), access to NVIDIA GPU Performance
Counters must be enabled in the NVIDIA Control Panel of the Windows host.
‣ Unsupported GPU
This indicates that the GPU, on which the current kernel is launched, is not
supported. See the Release Notes for a list of devices supported by your version of
NVIDIA Nsight Compute. It can also indicate that the current GPU configuration is
not supported. For example, NVIDIA Nsight Compute might not be able to profile
GPUs in SLI configuration.
‣ Connection error detected communicating with target application.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 68
 FAQ

The inter-process connection to the profiled application unexpectedly dropped. This

happens if the application is killed or signals an exception (e.g. segmentation fault).
‣ Failed to connect. The target process may have exited.
This occurs if
‣ the application does not call any CUDA API calls before it exits.
‣ the application terminates early because it was started from the wrong working
directory, or with the wrong arguments. In this case, check the details in the
Connection Dialog.
‣ the application crashes before calling any CUDA API calls.
‣ the application launches child processes which use the CUDA. In this case,
launch with the --target-processes all option.
‣ The profiler returned an error code: (number)
For the non-interactive Profile activity, the NVIDIA Nsight Compute CLI is started
to generate the report. If either the application exited with a non-zero return code,
or the NVIDIA Nsight Compute CLI encountered an error itself, the resulting return
code will be shown in this message.
For example, if the application hit a segmentation fault (SIGSEGV) on Linux, it will
likely return error code 11. All non-zero return codes are considered errors, so the
message is also shown if the application exits with return code 1 during regular
execution.
To debug this issue, it can help to run the data collection directly from the command
line using ncu in order to observe the application's and the profiler's command line
output, e.g. ==ERROR== The application returned an error code (11)
‣ Failed to open/create lock file (path). Please check that this process has write
permissions on this file.
NVIDIA Nsight Compute failed to create or open the file (path) with write
permissions. This file is used for inter-process serialization. NVIDIA Nsight
Compute does not remove this file after profiling by design. The error occurs if the
file was created by a profiling process with permissions that prevent the current
process from writing to this file, or if the current user can't acquire this file for other
reasons (e.g. certain Linux kernel security settings).
The file is in the current temporary directory, i.e. TMPDIR/nsight-compute-
lock. On Windows, TMPDIR is the path returned by the Windows GetTempPath
API function. On other platforms, it is the path supplied by the first environment
variable in the list TMPDIR, TMP, TEMP, TEMPDIR. If none of these is found, it's /
var/nvidia on QNX and /tmp otherwise.
Older versions of NVIDIA Nsight Compute did not set write permissions for all
users on this file by default. As a result, running the tool on the same system with a
different user might cause this error. This has been resolved since version 2020.2.1.
The following workarounds can be used to solve this problem:

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 69
 FAQ

‣ If it is otherwise ensured that no concurrent NVIDIA Nsight Compute instances

are active on the same system, set TMPDIR to a different directory for which the
current user has write permissions.
‣ Ask the user owning the file, or a system administrator, to remove it or add
write permissions for all potential users.
‣ On Linux systems setting fs.protected_regular=1, root or other users may
not be able to access this file, even though the owner can, if the sticky bit is
set on the temporary directory. Either disable this setting using sudo sysctl
fs.protected_regular=0, use a different temporary directory (see above),
or enable access to hardware performance counters for non-root users and
profile as the same user who owns the file (see https://developer.nvidia.com/
ERR_NVGPUCTRPERM on how to change this setting).
‣ Profiling failed because a driver resource was unavailable.
The error indicates that a required CUDA driver resource was unavailable during
profiling. Most commonly, this means that NVIDIA Nsight Compute could not
reserve the driver's performance monitor, which is necessary for collecting most
metrics.
This can happen if another application has a concurrent reservation on this resource.
Such applications can be e.g. DCGM, a client of CUPTI's Profiling API, Nsight
Graphics, or another instance of NVIDIA Nsight Compute without access to the
same file system (see serialization for how this is prevented within the same file
system).
If you expect the problem to be caused by DCGM, consider using dcgmi profile
--pause to stop its monitoring while profiling with NVIDIA Nsight Compute.
‣ Could not deploy stock * files to *
Could not determine user home directory for section deployment.
An error occurred while trying to deploy stock section or rule files. By default,
NVIDIA Nsight Compute tries to deploy these to a versioned directory in the user's
home directory (as identified by the HOME environment variable on Linux), e.g. /
home/user/Documents/NVIDIA Nsight Compute/<version>/Sections.
If the directory cannot be determined (e.g. because this environment variable is
not pointing to a valid directory), or if there is an error while deploying the files
(e.g. because the current process does not have write permissions on it), warning
messages are shown and NVIDIA Nsight Compute falls back to using stock sections
and rules from the installation directory.
If you are in an environment where you consistently don't have write access to the
user's home directory, consider populating this directory upfront using ncu --
section-folder-restore, or by making /home/user/Documents/NVIDIA
Nsight Compute/<version> a symlink to a writable directory.
‣ ProxyJump SSH option is not working
NVIDIA Nsight Compute does not manage authentication or interactive prompts
with the OpenSSH client launched when using the ProxyJump option. Therefore,
to connect through an intermediate host for the first time, you will not be able to

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 70
 FAQ

accept the intermediate host's key. A simple way to pinpoint the cause of failures in
this case is to open a terminal and use the OpenSSH client to connect to the remote
target. Once that connection succeeds, NVIDIA Nsight Compute should be able to
connect to the target, too.
‣ SSH connection fails without trying to connect
If the connection fails without trying to connect, there may be a problem with the
settings you entered into the connection dialog. Please make sure that the IP/Host
Name, User Name and Port fields are correctly set.
‣ SSH connections are still not working
The problem might come from NVIDIA Nsight Compute's SSH client not finding a
suitable host key algorithm to use which is supported by the remote server. You can
force NVIDIA Nsight Compute to use a specific set of host key algorithms by setting
the HostKeyAlgorithms option for the problematic host in your SSH configuration
file. To list the supported host key algorithms for a remote target, you can use the
ssh-keyscan utility which comes with the OpenSSH client.
‣ Removing host keys from known hosts files
When connecting to a target machine, NVIDIA Nsight Compute tries to verify the
target's host key against the same local database as the OpenSSH client. If NVIDIA
Nsight Compute find the host key is incorrect, it will inform you through a failure
dialog. If you trust the key hash shown in the dialog, you can remove the previously
saved key for that host by manually editing your known hosts database or using the
ssh-keygen -R <host> command.
‣ Qt initialization failed
Failed to load Qt platform plugin
See System Requirements for Linux.

www.nvidia.com
Kernel Profiling Guide v2023.1.1 | 71
Notice
ALL NVIDIA DESIGN SPECIFICATIONS, REFERENCE BOARDS, FILES, DRAWINGS,
DIAGNOSTICS, LISTS, AND OTHER DOCUMENTS (TOGETHER AND SEPARATELY,
"MATERIALS") ARE BEING PROVIDED "AS IS." NVIDIA MAKES NO WARRANTIES,
EXPRESSED, IMPLIED, STATUTORY, OR OTHERWISE WITH RESPECT TO THE
MATERIALS, AND EXPRESSLY DISCLAIMS ALL IMPLIED WARRANTIES OF
NONINFRINGEMENT, MERCHANTABILITY, AND FITNESS FOR A PARTICULAR
PURPOSE.
Information furnished is believed to be accurate and reliable. However, NVIDIA
Corporation assumes no responsibility for the consequences of use of such
information or for any infringement of patents or other rights of third parties
that may result from its use. No license is granted by implication of otherwise
under any patent rights of NVIDIA Corporation. Specifications mentioned in this
publication are subject to change without notice. This publication supersedes and
replaces all other information previously supplied. NVIDIA Corporation products
are not authorized as critical components in life support devices or systems
without express written approval of NVIDIA Corporation.

Trademarks
NVIDIA and the NVIDIA logo are trademarks or registered trademarks of NVIDIA
Corporation in the U.S. and other countries. Other company and product names
may be trademarks of the respective companies with which they are associated.

Copyright
© 2018-2023 NVIDIA Corporation and affiliates. All rights reserved.

This product includes software developed by the Syncro Soft SRL (http://
www.sync.ro/).

www.nvidia.com