The AMD ROCm™ 5 Open

Platform for HPC and ML

Workloads

Free Your Workloads With the ROCm™ 5 Platform

Whatever your workload, the AMD ROCm open software platform opens doors to new What’s New in ROCm 5?
levels of freedom and accessibility. When you use ROCm, you can run your software
on the GPU accelerator, the CPU, and the server platform of your choice. You only have 1. Expanded support and access:
to write your software once. Then you can test and debug it on a workload accelerator support for AMD Instinct™ MI210
and graduate to data center GPUs when the software is ready to tackle large problems. and AMD Radeon™ Pro W6800
You can seamlessly switch between accelerators from different vendors, and even workstation GPUs
accommodate different inter-GPU connectivity architectures—all in a device- 2. Optimized performance: new and
independent manner. improved FP64 matrix operations
and better cache handling, along
Accelerate Your High Performance Computing Workloads with improved kernel launch
Some of the most popular HPC frameworks are part of the ROCm platform, including latency and runtime
those to help parallelize operations across multiple GPUs and servers, handle
3. Enabling developer success:
memory hierarchies, and solve linear systems. Our GPU Accelerated Applications
prepackaged HPC and AI/ML
Catalog includes a vast set of platform-compatible HPC applications, including those
frameworks ready for download on
in astrophysics, climate & weather, computational chemistry, computational fluid
the AMD Infinity Hub; streamlined
dynamics, earth science, genomics, geophysics, molecular dynamics, and physics.
and improved tools
Many of these are available through the AMD Infinity Hub, ready to download and run
on servers with AMD Instinct™ accelerators. 4. Easy access to resources: new
ROCm information portal, AMD
Propel Your Machine Learning Applications community support, and remote
Upstream support for the most popular ML frameworks—TensorFlow , Caffe, and access to AMD Accelerator Cloud
PyTorch—make it easy to adopt the ROCm platform. A broad range of supporting (AAC) for development, test, and
libraries to help improve your models and run-time environments including benchmarking
DeepSpeed, ONNX-RT, Jax, and CuPy all are supported as part of the ROCm platform.
The AMD Infinity Hub provides ready-to-run containerized frameworks, and our GPU “One of the benefits
Accelerated Applications Catalog lists the broad set of supporting libraries that are
available to AI and ML practitioners. of converting to HIP
is that unlike previous
Kokkos OpenMPI CUDA versions of code
Raja MVAPICH2
AMReX UCX that could only run
Legion RCCL
Alpaka SHMEM LLVM on Nvidia GPUs, the
TensorFlow GCC
PyTorch
DeepSpeed
HPC Scale-Out
Frameworks Computing
HIP
OpenMP
same source code is
ONNX-RT
Jax GDB
now portable between
ML PAPI
cuPy
Frameworks Toolchain
JSON trace GPUs.”
Graphana Deployment & Linux Kernel driver Daniel Jacobson, Computational
Prometheus Management Distributions .rpm packages
Kubernetes plug-in Tools .deb packages
Systems Biologist, ORNL
OCI (Containers) Spack

Copyright 2022 Advanced Micro Devices, Inc. All Rights Reserved.

©2019 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, Radeon, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Thunderbolt is a trademark of Intel Corporation
or its subsidiaries in the U.S. and/or other countries. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.
 AMD ROCm™ 5 FOR HPC AND ML WORKLOADS

The ROCm Stack

Regardless of your workload, the ROCm stack includes the deployment
and management tools, optimized libraries, and programming and
system tools you need for compiling, debugging, performance analysis,
and system management:
• Deployment & Management Tools simplify the process of deploying
and running your HPC and ML codes, including a validation suite
that ensures that your environment is ready to host your software;
the ROCm Data Center Tool™ helps gather telemetry and statistics
Development & Management Tools for your jobs, including integration with third-party tools; and
even monitoring environmental factors such as temperature are
Application Frameworks supported with the AMD System Management Interface
• Key Industry and Application Frameworks include support for HPC
Libraries and ML software as discussed above.
• Libraries include support for math functions, distributed computing,
Programming Models containers, and scale-out communication.
• Programming Models include support for OpenMP®, HIP, and
Device Drivers & Runtimes OpenCL™, and tools to help you compile and run, profile, and debug
your software. ROCm supports C/C++ and provides tools that can
GPU Support automatically make your CUDA software universal by converting it
to HIP, the Heterogeneous Compute Interface for Portability.
• Device driver and runtime environments support Red Hat®
Enterprise Linux, SUSE Linux Enterprise Server Distribution, and
Ubuntu® Linux. One of the benefits of ROCm is that it is easy for
vendors to create device drivers for their accelerators, extending the
platform’s range and diversity to include both workstation and data-
center class GPU accelerators.
• GPU Support includes a broad range of AMD Radeon and Instinct
accelerators along with being open to support third-party GPU and
FPGA devices

Learn More
We’ve made the ROCm 5 platform even easier to learn about, install, and run:
• ROCm Information Portal posts the latest versions of ROCm along with API and support documentation, curated videos, webinars,
labs, and tutorials to help you learn how to use the platform. It replaces the former documentation and learning sites.
• AMD Infinity Hub gives you access to HPC applications and ML frameworks packaged as containers and ready to run. We’ve
expanded our already big catalog to include weather, QFD, quantum chemistry, and more.
• ROCm Application Catalog gives the most full and up-to-date list of applications supported by the ROCm platform.
• AMD ROCm Community is an excellent place access real-time discussion support, knowledge bases, and the latest news through
blogs on the ROCm open software platform.
• AMD Accelerator Cloud offers remote access to test code and applications in the cloud, on the latest AMD GPUs, including the
Instinct MI200 series

For More Information Visit AMD.com/ROCm

© 2022 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow, Instinct, Radeon, ROCm, EPYC, and combinations thereof are trademarks of Advanced Micro Devices, Inc. PCIe® is a registered trademark of
PCI-SIG Corporation. SPEC®, SPECrate® and SPEC CPU® are registered trademarks of the Standard Performance Evaluation Corporation. See www.spec.org for more information. Other trademarks used in this publication
are for identification purposes only and may be trademarks of their respective companies. LE-82107-00 04/22

Copyright 2022 Advanced Micro Devices, Inc. All Rights Reserved.

©2019 Advanced Micro Devices, Inc. All rights reserved. AMD, the AMD Arrow logo, Radeon, and combinations thereof are trademarks of Advanced Micro Devices, Inc. Thunderbolt is a trademark of Intel Corporation
or its subsidiaries in the U.S. and/or other countries. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.