Scribd
Upload
22 views3 pages

Parallel 4

Isie aass ssss sass s ss s s

Uploaded by

Rooman M
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
22 views3 pages

Parallel 4

Isie aass ssss sass s ss s s

Uploaded by

Rooman M
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Detailed Note on Components of GPU

A Graphics Processing Unit (GPU) is an essential component in modern computing,


designed primarily for rendering images and videos. Beyond graphics, GPUs are
widely used in scientific computations, machine learning, and other data-intensive
tasks due to their highly parallel architecture. Below is a detailed explanation of the
key components that make up a GPU:

1. Graphics Processing Clusters (GPCs)

Graphics Processing Clusters (GPCs) are fundamental building blocks within a GPU.
Each GPC includes multiple Streaming Multiprocessors (SMs) and other essential
units for graphics rendering. GPCs manage high-level tasks such as vertex processing,
geometry shading, and pixel shading.

2. Streaming Multiprocessors (SMs)

Streaming Multiprocessors are the core units that perform the bulk of computational
tasks. Each SM includes:

 CUDA Cores/Stream Processors: The primary execution units responsible


for performing arithmetic and logical operations. CUDA cores are used in
NVIDIA GPUs, while AMD GPUs use Stream Processors.
 Warp/Wavefront Scheduler: Manages the execution of groups of threads
(called warps in NVIDIA GPUs and wavefronts in AMD GPUs).
 Register File: Stores temporary data for quick access by CUDA cores.
 Shared Memory/Local Data Store: On-chip memory that allows threads
within the same block to share data quickly.
 Texture Units: Specialized units for fetching and processing texture data.

3. Memory Hierarchy

The memory hierarchy in a GPU is designed to maximize data access speed and
bandwidth:

 Global Memory: The largest but slowest memory type, accessible by all
threads.
 Local Memory: Memory specific to a thread, used for storing private
variables.
 Shared Memory: Fast, on-chip memory shared by all threads within a block.
 L1/L2/L3 Caches: Multi-level caches that store frequently accessed data to
reduce access latency.
 Texture Memory: Specialized memory optimized for texture data retrieval
and processing.

4. Raster Operators (ROPs)

Raster Operators, also known as Render Output Units (ROPs) or Pixel Processors,
handle tasks like pixel blending, antialiasing, and writing final pixel data to the frame
buffer. ROPs are crucial in the final stage of the rendering pipeline, processing the
output of pixel shaders.

5. Texture Mapping Units (TMUs)

Texture Mapping Units (TMUs) fetch and filter texture data. They take texture
coordinates, fetch the corresponding texture data from memory, and apply filtering
techniques to produce the final texture color used in rendering.

6. Shader Units

Shaders are programmable units that allow custom processing of vertex, geometry,
and pixel data. Types of shaders include:

 Vertex Shaders: Process vertex data for tasks like transformations and lighting.
 Geometry Shaders: Process geometric primitives such as points, lines, and triangles.
 Pixel/Fragment Shaders: Process pixel data to determine the final color of each pixel.

7. Command Processor

The Command Processor interprets and executes commands from the CPU, managing
the GPU's overall operation, including task scheduling, context switching, and
synchronization.

8. Memory Controller

The Memory Controller handles data transfers between the GPU and its memory,
managing memory requests from different GPU parts to optimize access patterns and
improve performance.

9. Interconnects and Buses

Interconnects and buses facilitate communication within the GPU and between the
GPU and other system components, such as the CPU and system memory. Common
interconnects include:

 PCIe (Peripheral Component Interconnect Express): The interface for communication


between the GPU and the CPU.
 NVLink: A high-speed interconnect developed by NVIDIA for connecting multiple GPUs
and enhancing data transfer rates.

10. Display Controllers

Display Controllers manage the output of rendered images to display devices,


handling tasks like resolution scaling, refresh rate control, and multiple display
outputs.

11. Cooling System


The Cooling System, although not a computational component, is crucial for
maintaining the GPU's performance and longevity. It includes heat sinks, fans, and
sometimes liquid cooling solutions to dissipate heat generated during operation.

You might also like