* Update RELEASE.md and 5.7.0.md to match CHANGELOG.md
* Update 5.2.0.md to match CHANGELOG.md
* Copy CHANGELOG into about folder to match RELEASE
To avoid having divergence in relative links between RELEASE and CHANGELOG
Update docs with information in the AMD blog post announcing support for some RDNA3 Radeon GPUs on Linux.
Co-authored-by: Saad Rahim (AMD) <44449863+saadrahim@users.noreply.github.com>
dependabot mis-detected the repository to be a library
(instead of an application) and widened the rocm-docs-core verison
instead of increasing it. This basically disabled pinning.
Explicitly specify to increase the version instead of widening it
to hopefully prevent this in the future.
Flattened out page structure for improved navigability.
* Change Table of Contents
* Update the install guides for windows and linux
* Removed extraneous index pages
* GPU architecture pages duplicate entries removed
* spack page cleanup
---------
Co-authored-by: Sam Wu <samwu103@amd.com>
Co-authored-by: Saad Rahim (AMD) <44449863+saadrahim@users.noreply.github.com>
Since we are not installing the ROCm OpenCL packages. We are not able to
test ROCm withg this command.
Signed-off-by: Tasso Zambelakis <Tasso.Zambelakis@amd.com>
* RX 6700* doc fixes in windows_support.md
Correct RX 6700* LLVM target to gfx1031 windows_support.md
Change name from "RX 6750" to "RX 6750 XT"
* Fix RX7600 LLVM to gfx1102 in windows-support.md
---------
Co-authored-by: Saad Rahim (AMD) <44449863+saadrahim@users.noreply.github.com>
This should increase usability and prevent errors, since the most common
use case is the user using the latest version of their OS,
rather than the oldest supported one.
* update relative link to llvm asan guide
remove docs dir from path
* Minor typo and update on supported OSes
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* Update install instructions to 5.7
* RTG additions to install instructions
* update install instructions for multi version
---------
Co-authored-by: Máté Ferenc Nagy-Egri <mate@streamhpc.com>
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* Update using-gpu-sanitizer.md
Updated content
* fixes for markdown linting
use * instead of + for lists
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* Create using_gpu_sanitizer.md
* Created GPU Sanitizer File and Title
* add technical terms to wordlist and fix spelling
* spelling
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
Co-authored-by: b-sumner <brian.sumner@amd.com>
* Added deleted sections to openmp.md and other improvements
* Update CONTRIBUTING.md
* add example of snake case
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* spell out HPC acronym in explanation doc
* update toolchain docs
order in importance descending
* update Contributing guide
add discussions
update formatting and grammar
* separate contributing section for readability
* fix formatting for mdl
* fix spelling
* Update Links (#2240)
* update link to PCIe Gen 4 pdf
* fix broken links
* remove references to broken links
* fix spelling of data center
* Fixing HIP link (#2236)
* Swati develop (#2245)
* Added deleted sections to openmp.md and other improvements
* Update openmp.md
Tagged `ICV`
* Solving indiscrepencies in openmp.md
There are apparently differences in the published document and information conveyed by the Dev. Fixed it.
* add new words to wordlist
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* fix rocm_smi_lib link in toc (#2260)
* ROCm FHS Reorganization, Backward Compatibility, and Versioning - rev (#2255)
* update requirements
---------
Co-authored-by: Saad Rahim <44449863+saadrahim@users.noreply.github.com>
Co-authored-by: srawat <120587655+SwRaw@users.noreply.github.com>
Co-authored-by: Ehud Sharlin <112672820+Ehud-Sharlin@users.noreply.github.com>
* docs: update changelog and release notes with hipStreamGetDevice
* docs: fix typos and add version update notes
* docs: add HIP changelog
* remove What's New section from changelog
* docs: clean up SLES tab-sets
- Always use a tab-set for SLES 15.4
- In the toplevel SLES title don't say version 15
- harmonize the `:sync:` labels between documents
* docs: Misc fixes in installation
- Fix rocm repository url in the installer script installation for SLES
- Add a missing :sync: tab in installation prerequisites
* docs: add SLES 15.5 support to installation and OS support pages
* Added deleted sections to openmp.md and other improvements
* Update openmp.md
Tagged `ICV`
* Solving indiscrepencies in openmp.md
There are apparently differences in the published document and information conveyed by the Dev. Fixed it.
* add new words to wordlist
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* Update tensorflow_install.md
Fixed writing commands so that when executed by copy paste it doesn't cause an error.
* Update tensorflow_install.md
Following @saadrahim's suggestion of using "\" to signify a line break in bash.
* Remove package pin from quick start quide
When installing a single-package fashion, no version pinning is needed
* Add package pinning to quick start guide
Pinning the packages is required to make apt prefer the rocm packages
instead of the system ones when both provide the same package (e.g
`rocm-smi`).
* Removing Ubuntu 20.04 change
---------
Co-authored-by: Gergely Meszaros <gergely@streamhpc.com>
Co-authored-by: Saad Rahim <44449863+saadrahim@users.noreply.github.com>
* Added specialized kernels to openmp.md
A few formatting changes and addition of specialized kernels section at the end.
* Added Specialized kernels in openmp.md
Some formatting changes and addition of specialized kernels instead of no loop and cross team kernels
* Added specialized kernel to openmp.md
* Added specialized kernels to openmp.md
* Replaced the usage of uncertain clauses(may/might) in openmp.md
* Attempt to align the table headings for environment variables in openmp.md
* Feedback from Dhruva
---------
Co-authored-by: Saad Rahim <44449863+saadrahim@users.noreply.github.com>
The hip-devel package depends on perl modules not distributed by default
on RHEL and SLES distriubutions, these can be installed from EPEL and
the `devel:languages:perl` repository respectively.
Ideally in the future these dependencies would be replaced with packages
available from default repositories, but in the meanwhile this should
be at least documented.
* Remove install instructions for unsuported RHEL 8.8 and 9.2
Current ROCm release does not support these versions of RHEL
* Centralize disclaimers and perquisites for installation
- Move the single-version to multi-version diclaimer to the install
overview page where single vs multi installs are discussed.
- Move the installation of kernel-headers and development packages
to the install preparation page. Unify it mainly from the quick start
content.
* s/Name/name/ in repository config files for RHEL
The repository name can be set as `name=><name>` instead of `Name`,
otherwise yum complains about the repo not having a name, e.g:
```output
Repository 'ROCm-5.3.3' is missing name in configuration, using id.
```
This is fixed with this commit.
* Clean up render/video group section on prerequisites
* Installation and Upgrade restructuring & fixes
- Fix the rocm package urls for RHEL in the install & upgrade guides
- RHEL8 and 9 have different URLs, add a tab-set similar to ubuntu
for them.
- Fix the package URL in the upgrade guide for SLES (previously pointed
to the amdgpu url)
- Change the apt-signing key download and conversion to the method used
in the quick start guide, which is the recommended by ubuntu maintainers
- Change the install steps from list items to rubrics with numbered entries
which is more readable and matches the style in the quick start guide
- Do not pass `--append` to `tee` in the upgrade guide, because it is
meant to overwrite.
- Split the one long tab-set to multiple tab-sets in the upgrade guide
to improve readability
Add empty cells to list tables to make them uniform (all rows have the
same number of cells), before this the tables errored out with:
> ERROR: Error parsing content block for the "list-table" directive:
> uniform two-level bullet list expected, but row 13 does not contain
> the same number of items as row 1 (3 vs 4)
and the table did not show up.
* ci: change markdown linting to use the NodeJs markdownlint
The original ruby based markdownlint has a few shortcomings not known
when it was introduced:
- no support for myst extensions
- no support for disabling specific rules for specific files or regions
These two combined make it very hard to use when used for this project
when it has false positives around myst extensions.
Luckily there's a NodeJS based version of markdownlint [1] supporting the
same ruleset that is more configurable:
- seems to support myst extensions better
- has an html comment based syntax to disable specific rules
The library seem to be better maintained too and with better tooling:
e.g. there's a vscode extension using the engine for local use:
markdownlint (DavidAnson.vscode-markdownlint).
[1]: https://github.com/DavidAnson/markdownlint
* docs: hotfix empty links
There are missing links in the docs, these should get fixed, but for now
they are just monkey patched to make CI happy.
* docs: fix links
---------
Co-authored-by: Nara Prasetya <nara@streamhpc.com>
* update the gpu-aware-mpi page
Three changes:
- add the ucx compatibility table
- add the --with-rocm=/opt/rocm option to the compilation of Open MPI
- add a section about how to compile and use UCC for collective
operations.
* Changing link to relative
* Update gpu_aware_mpi.md
---------
Co-authored-by: Saad Rahim <44449863+saadrahim@users.noreply.github.com>
* fix rocmcc link
* remove unused link
* remove unused linkcheck configs
* update amd smi section
add link to ami smi github
---------
Co-authored-by: Saad Rahim <44449863+saadrahim@users.noreply.github.com>
* update links to new docs and rename .sphinx dir to sphinx
* fix spelling and formatting
add new words to wordlist
remove empty headers
remove version number for ROCm in conf.py
fix typos
* add more formats to rtd config
* Clean up the deployment related pages
- Add an index page for the linux deployment submenu
- Remove deployment options that are not yet completed (i.e. spack,
from source installation)
- remove the general deployment index page
- various cleanups and clarifications in the rest of the pages
* Move all deploy pages to deploy folder
---------
Co-authored-by: Gergely Meszaros <gergely@streamhpc.com>
- Unify code block style (indent vs. fence)
- Mark code languages
- Increase heading level one at a time
- No extra newlines between paragraphs
- List for header reorg stages
- Shrink ascii table (mobile friendlyness)
- 80-column width
* add url to ROCgdb-docs
update reqs and gitignore
* add validation tools section for RVS and TransferBench
* stub in links for validation/mgmt tools
* populate compilers page
* add cards for ai libs and computer vision pages
* add content to math lib pages
* reorg hip and math libs
* update index
* consolidate linear algebra libs
* fix release info order in toc
* fix links and content cards for libraries
* update mdl ignored files
* update understand rocm section
* fix formatting errors
* add link to openmp
* ignore md041
* add url to ROCgdb-docs
update reqs and gitignore
* add validation tools section for RVS and TransferBench
* stub in links for validation/mgmt tools
* populate compilers page
* add cards for ai libs and computer vision pages
* add content to math lib pages
* reorg hip and math libs
* update index
* consolidate linear algebra libs
* Pulling libraries out
* add libraries listed in left sidebar to index page
* Adding all
* Updating nav tree
* fix link to rocm-examples in toc
* update TOC
---------
Co-authored-by: Sam Wu <sam.wu2@amd.com>
* Add C++ algorithm primitive lib cards
* Add PRNG section
* API Reference Manuals first
* Add Tensile and rocWMMA
* Change rocFFT and hipFFT order for consistency
* Add RCCL
* Fix PRNG links
* Add rocSOLVER and hipSOLVER
* Add general note on rocLIB vs hipLIB
* linux quick start: Mention correct package to install
* linux quick start: Rephrase prerequisites
Mention that installing the headers is usually not required by hand.
* linux quick start: Simplify command to get singing key
* linux quick start: Add instructions for RHEL and SLES
Reorganize the quick start guide for linux, adding multi level
tab selection for just the commands where it makes sense.
Currently mostly Ubuntu commands are filled out, if the structure
looks fine, then more will follow.
Add more of the sphinx generated files, so generating the docs does not
add untracked files. Ignore the folder `.venv` typically used for
virtual environments.
Also sort the ignored file list for easier maintenance.
The screenshots are from tables with text, which are not easily searchable,
are bigger in size than needed – increasing load times – and are in a
resolution, causing them to be blurry on HiDPI displays. Therefore, use a
Markdown table instead solving all the issues above, and delete the images
from the repository.
The SLES service pack version differs in the two screenshots: SP2 vs SP3.
Go for *SP3*.
Resolves: https://github.com/RadeonOpenCompute/ROCm/issues/1591
This document describes the features, fixed issues, and information about downloading and installing the AMD ROCm™ software. It also covers known issues and deprecations in this release.
- [Supported Operating Systems and Documentation Updates](#Supported-Operating-Systems-and-Documentation-Updates)
- [What\'s New in This Release](#Whats-New-in-This-Release)
* [HIP Enhancements](#HIP-Enhancements)
* [ROCm Data Center Tool](#ROCm-Data-Center-Tool)
* [ROCm Math and Communication Libraries](#ROCm-Math-and-Communication-Libraries)
- [Fixed Defects](#Fixed-Defects)
- [Known Issues](#Known-Issues)
- [Deprecations](#Deprecations)
* [Compiler Generated Code Object Version 2 Deprecation ](#Compiler-Generated-Code-Object-Version-2-Deprecation)
- [Deploying ROCm](#Deploying-ROCm)
- [Hardware and Software Support](#Hardware-and-Software-Support)
- [Machine Learning and High Performance Computing Software Stack for AMD GPU](#Machine-Learning-and-High-Performance-Computing-Software-Stack-for-AMD-GPU)
* CentOS 8.3 (4.18.0-193.el8)and RHEL 8.3 (4.18.0-193.1.1.el8) (devtoolset is not required)
* SLES 15 SP2
### Complete Installation of AMD ROCM V4.2 Recommended
Complete uninstallation of previous ROCm versions is required before installing a new version of ROCm. **An upgrade from previous releases to AMD ROCm v4.2 is not supported**. For more information, refer to the AMD ROCm Installation Guide at
**Note**: AMD ROCm release v3.3 or prior releases are not fully compatible with AMD ROCm v3.5 and higher versions. You must perform a fresh ROCm installation if you want to upgrade from AMD ROCm v3.3 or older to 3.5 or higher versions and vice-versa.
**Note**: *render* group is required only for Ubuntu v20.04. For all other ROCm supported operating systems, continue to use video group.
* For ROCm v3.5 and releases thereafter, the clinfo path is changed to /opt/rocm/opencl/bin/clinfo.
* For ROCm v3.3 and older releases, the clinfo path remains /opt/rocm/opencl/bin/x86_64/clinfo.
## ROCm Multi-Version Installation Update
With the AMD ROCm v4.2 release, the following ROCm multi-version installation changes apply:
The meta packages rocm-dkms<version> are now deprecated for multi-version ROCm installs. For example, rocm-dkms3.7.0, rocm-dkms3.8.0.
* Multi-version installation of ROCm should be performed by installing rocm-dev<version> using each of the desired ROCm versions. For example, rocm-dev3.7.0, rocm-dev3.8.0, rocm-dev3.9.0.
* The rock-dkms loadable kernel modules should be installed using a single rock-dkms package.
* ROCm v3.9 and above will not set any ldconfig entries for ROCm libraries for multi-version installation. Users must set LD_LIBRARY_PATH to load the ROCm library version of choice.
**NOTE**: The single version installation of the ROCm stack remains the same. The rocm-dkms package can be used for single version installs and is not deprecated at this time.
### Updated HIP Instructions for ROCm Installation
The hip-base package has a dependency on Perl modules that some operating systems may not have in their default package repositories. Use the following commands to add repositories that have the required Perl packages:
The platform macros are updated to target either the AMD or NVIDIA platform in HIP projects. They now include corresponding headers and libraries for compilation/linking.
* *__HIP_PLATFORM_AMD__* is defined if the HIP platform targets AMD. Note, __HIP_PLATFORM_HCC__ was used previously if the HIP platform targeted AMD.
This is now deprecated.
* *__HIP_PLATFORM_NVIDIA__* is defined if the HIP platform targets NVIDIA. Note, _HIP_PLATFORM_NVCC__ was used previously if the HIP platform targeted NVIDIA. This is now deprecated.
In the ROCm4.2 release, HIP *include* header directories for platforms are updated as follows:
* *amd_detail/* - includes source header details for the ‘amd’ platform implementation. In previous releases, the "hcc_detail" directory was defined, and it it is now deprecated.
* *nvidia_detail/* - includes source header details for the ‘nvidia’ platform implementation. In previous releases, the "nvcc_detail" directory was defined, and it is now deprecated.
### HIP Stream Memory Operations
The ROCm v4.2 extends support to Stream Memory Operations to enable direct synchronization between Network Nodes and GPU. The following new APIs are added:
HIP events in kernel dispatch using *hipExtLaunchKernelGGL/hipExtLaunchKernel* and passed in the API are not explicitly recorded and should only be used to get elapsed time for that specific launch.
Events used across multiple dispatches, for example, start and stop events from different *hipExtLaunchKernelGGL/hipExtLaunchKernel* calls, are treated as invalid unrecorded events. In such scenarios, HIP will display the error *"hipErrorInvalidHandle"* from *hipEventElapsedTime*.
In the ROCm v3.5 release, the Heterogeneous Compute Compiler (HCC) compiler was deprecated, and the HIP-Clang compiler was introduced for compiling Heterogeneous-Compute Interface for Portability (HIP) programs. In addition, the HIP runtime API was implemented on top of Radeon Open Compute Common Language Runtime (ROCclr). ROCclr is an abstraction layer that provides the ability to interact with different runtime backends such as ROCr.
While the HIP_PLATFORM=hcc environment variable was functional in subsequent releases, in the ROCm v4.1 release, the following environment variables were changed:
* *HIP_PLATFORM=hcc to HIP_PLATFORM=amd*
* *HIP_PLATFORM=nvcc to HIP_PLATFORM=nvidia*
Therefore, any applications continuing to use the HIP_PLATFORM=hcc variable will fail. You must update the environment variables to reflect the changes as mentioned above.
## ROCm Data Center Tool
### RAS Integration
The ROCm Data Center (RDC) Tool is enhanced with the Reliability, Accessibility, and Serviceability (RAS) plugin.
For more information about RAS integration and installation, refer to the ROCm Data Center Tool User guide at:
The following issue is fixed in the ROCm v4.2 release.
The compiler in ROCm v4.1 generates LDS load and stores instructions that incorrectly assume equal performance between aligned and misaligned accesses. While this does not impact code correctness, it may result in sub-optimal performance.
# Known Issues
The following are the known issues in this release.
## Upgrade to AMD ROCm v4.2 Not Supported
An upgrade from previous releases to AMD ROCm v4.2 is not supported. Complete uninstallation of previous ROCm versions is required before installing a new version of ROCm.
The hip-base package has a dependency on Perl modules that some operating systems may not have in their default package repositories. Use the following commands to add repositories that have the required Perl packages:
## Modulefile Fails to Install Automatically in ROCm Multi-Version Environment
The ROCm v4.2 release includes a preliminary implementation of environment modules to enable switching between multi versions of ROCm installation. The modulefile in */opt/rocm-4.2/lib/rocmmod* fails to install automatically in the ROCm multi-version environment.
This is a known limitation for environment modules in ROCm, and the issue is under investigation at this time.
**Workaround**
Ensure you install the modulefile in */opt/rocm-4.2/lib/rocmmod* manually in a multi-version installation environment.
For general information about modules, see
http://modules.sourceforge.net/
## Issue with Input/Output Types for Scan Algorithms in rocThrust
As rocThrust is updated to match CUDA Thrust 1.10, the different input/output types for scan algorithms in rocThrust/CUDA Thrust are no longer officially supported. In this situation, the current C++ standard does not specify the intermediate accumulator type leading to potentially incorrect results and ill-defined behavior.
As a workaround, users can:
* Use the same types for input and output
Or
* For exclusive_scan, explicitly specify an *InitialValueType* in the last argument
Or
* For inclusive_scan, which does not have an initial value argument, use a transform_iterator to explicitly cast the input iterators to match the output’s value_type
## Precision Issue in AMD RADEON™ PRO VII and AMD RADEON™ VII
In AMD Radeon™ Pro VII AND AMD Radeon™ VII, a precision issue can occur when using the Tensorflow XLA path.
This issue is currently under investigation.
# Deprecations
This section describes deprecations and removals in AMD ROCm.
## Compiler Generated Code Object Version 2 Deprecation
Compiler-generated code object version 2 is no longer supported and has been completely removed. Support for loading code object version 2 is also deprecated with no announced removal release.
# Deploying ROCm
AMD hosts both Debian and RPM repositories for the ROCm packages.
For more information on ROCM installation on all platforms, see
ROCm is focused on using AMD GPUs to accelerate computational tasks such as machine learning, engineering workloads, and scientific computing.
In order to focus our development efforts on these domains of interest, ROCm supports a targeted set of hardware configurations which are detailed further in this section.
**Note:** The AMD ROCm™ open software platform is a compute stack for headless system deployments. GUI-based software applications are currently not supported.
#### Supported GPUs
Because the ROCm Platform has a focus on particular computational domains, we offer official support for a selection of AMD GPUs that are designed to offer good performance and price in these domains.
**Note:** The integrated GPUs of Ryzen are not officially supported targets for ROCm.
ROCm officially supports AMD GPUs that use following chips:
* GFX9 GPUs
- "Vega 10" chips, such as on the AMD Radeon RX Vega 64 and Radeon Instinct MI25
- "Vega 7nm" chips, such as on the Radeon Instinct MI50, Radeon Instinct MI60 or AMD Radeon VII, Radeon Pro VII
* CDNA GPUs
- MI100 chips such as on the AMD Instinct™ MI100
ROCm is a collection of software ranging from drivers and runtimes to libraries and developer tools.
Some of this software may work with more GPUs than the "officially supported" list above, though AMD does not make any official claims of support for these devices on the ROCm software platform.
The following list of GPUs are enabled in the ROCm software, though full support is not guaranteed:
* GFX8 GPUs
* "Polaris 11" chips, such as on the AMD Radeon RX 570 and Radeon Pro WX 4100
* "Polaris 12" chips, such as on the AMD Radeon RX 550 and Radeon RX 540
* GFX7 GPUs
* "Hawaii" chips, such as the AMD Radeon R9 390X and FirePro W9100
As described in the next section, GFX8 GPUs require PCI Express 3.0 (PCIe 3.0) with support for PCIe atomics. This requires both CPU and motherboard support. GFX9 GPUs require PCIe 3.0 with support for PCIe atomics by default, but they can operate in most cases without this capability.
The integrated GPUs in AMD APUs are not officially supported targets for ROCm.
As described [below](#limited-support), "Carrizo", "Bristol Ridge", and "Raven Ridge" APUs are enabled in our upstream drivers and the ROCm OpenCL runtime.
However, they are not enabled in the HIP runtime, and may not work due to motherboard or OEM hardware limitations.
As such, they are not yet officially supported targets for ROCm.
For a more detailed list of hardware support, please see [the following documentation](https://en.wikipedia.org/wiki/List_of_AMD_graphics_processing_units).
#### Supported CPUs
As described above, GFX8 GPUs require PCIe 3.0 with PCIe atomics in order to run ROCm.
In particular, the CPU and every active PCIe point between the CPU and GPU require support for PCIe 3.0 and PCIe atomics.
The CPU root must indicate PCIe AtomicOp Completion capabilities and any intermediate switch must indicate PCIe AtomicOp Routing capabilities.
Current CPUs which support PCIe Gen3 + PCIe Atomics are:
* AMD Ryzen CPUs
* The CPUs in AMD Ryzen APUs
* AMD Ryzen Threadripper CPUs
* AMD EPYC CPUs
* Intel Xeon E7 v3 or newer CPUs
* Intel Xeon E5 v3 or newer CPUs
* Intel Xeon E3 v3 or newer CPUs
* Intel Core i7 v4, Core i5 v4, Core i3 v4 or newer CPUs (i.e. Haswell family or newer)
* Some Ivy Bridge-E systems
Beginning with ROCm 1.8, GFX9 GPUs (such as Vega 10) no longer require PCIe atomics.
We have similarly opened up more options for number of PCIe lanes.
GFX9 GPUs can now be run on CPUs without PCIe atomics and on older PCIe generations, such as PCIe 2.0.
This is not supported on GPUs below GFX9, e.g. GFX8 cards in the Fiji and Polaris families.
If you are using any PCIe switches in your system, please note that PCIe Atomics are only supported on some switches, such as Broadcom PLX.
When you install your GPUs, make sure you install them in a PCIe 3.1.0 x16, x8, x4, or x1 slot attached either directly to the CPU's Root I/O controller or via a PCIe switch directly attached to the CPU's Root I/O controller.
In our experience, many issues stem from trying to use consumer motherboards which provide physical x16 connectors that are electrically connected as e.g. PCIe 2.0 x4, PCIe slots connected via the Southbridge PCIe I/O controller, or PCIe slots connected through a PCIe switch that does
not support PCIe atomics.
If you attempt to run ROCm on a system without proper PCIe atomic support, you may see an error in the kernel log (`dmesg`):
Experimental support for our Hawaii (GFX7) GPUs (Radeon R9 290, R9 390, FirePro W9100, S9150, S9170)
does not require or take advantage of PCIe Atomics. However, we still recommend that you use a CPU
from the list provided above for compatibility purposes.
#### Not supported or limited support under ROCm
##### Limited support
* ROCm 4.x should support PCIe 2.0 enabled CPUs such as the AMD Opteron, Phenom, Phenom II, Athlon, Athlon X2, Athlon II and older Intel Xeon and Intel Core Architecture and Pentium CPUs. However, we have done very limited testing on these configurations, since our test farm has been catering to CPUs listed above. This is where we need community support. _If you find problems on such setups, please report these issues_.
* Thunderbolt 1, 2, and 3 enabled breakout boxes should now be able to work with ROCm. Thunderbolt 1 and 2 are PCIe 2.0 based, and thus are only supported with GPUs that do not require PCIe 3.1.0 atomics (e.g. Vega 10). However, we have done no testing on this configuration and would need community support due to limited access to this type of equipment.
* AMD "Carrizo" and "Bristol Ridge" APUs are enabled to run OpenCL, but do not yet support HIP or our libraries built on top of these compilers and runtimes.
* As of ROCm 2.1, "Carrizo" and "Bristol Ridge" require the use of upstream kernel drivers.
* In addition, various "Carrizo" and "Bristol Ridge" platforms may not work due to OEM and ODM choices when it comes to key configurations parameters such as inclusion of the required CRAT tables and IOMMU configuration parameters in the system BIOS.
* Before purchasing such a system for ROCm, please verify that the BIOS provides an option for enabling IOMMUv2 and that the system BIOS properly exposes the correct CRAT table. Inquire with your vendor about the latter.
* AMD "Raven Ridge" APUs are enabled to run OpenCL, but do not yet support HIP or our libraries built on top of these compilers and runtimes.
* As of ROCm 2.1, "Raven Ridge" requires the use of upstream kernel drivers.
* In addition, various "Raven Ridge" platforms may not work due to OEM and ODM choices when it comes to key configurations parameters such as inclusion of the required CRAT tables and IOMMU configuration parameters in the system BIOS.
* Before purchasing such a system for ROCm, please verify that the BIOS provides an option for enabling IOMMUv2 and that the system BIOS properly exposes the correct CRAT table. Inquire with your vendor about the latter.
##### Not supported
* "Tonga", "Iceland", "Vega M", and "Vega 12" GPUs are not supported.
* We do not support GFX8-class GPUs (Fiji, Polaris, etc.) on CPUs that do not have PCIe 3.0 with PCIe atomics.
* As such, we do not support AMD Carrizo and Kaveri APUs as hosts for such GPUs.
* Thunderbolt 1 and 2 enabled GPUs are not supported by GFX8 GPUs on ROCm. Thunderbolt 1 & 2 are based on PCIe 2.0.
In the default ROCm configuration, GFX8 and GFX9 GPUs require PCI Express 3.0 with PCIe atomics. The ROCm platform leverages these advanced capabilities to allow features such as user-level submission of work from the host to the GPU. This includes PCIe atomic Fetch and Add, Compare and Swap, Unconditional Swap, and AtomicOp Completion.
#### ROCm support in upstream Linux kernels
As of ROCm 1.9.0, the ROCm user-level software is compatible with the AMD drivers in certain upstream Linux kernels.
As such, users have the option of either using the ROCK kernel driver that are part of AMD's ROCm repositories or using the upstream driver and only installing ROCm user-level utilities from AMD's ROCm repositories.
These releases of the upstream Linux kernel support the following GPUs in ROCm:
| Pros | More GPU features, and they are enabled earlier | Includes the latest Linux kernel features |
| | Tested by AMD on supported distributions | May work on other distributions and with custom kernels |
| | Supported GPUs enabled regardless of kernel version | |
| | Includes the latest GPU firmware | |
| Cons | May not work on all Linux distributions or versions | Features and hardware support varies depending on kernel version |
| | Not currently supported on kernels newer than 5.4 | Limits GPU's usage of system memory to 3/8 of system memory (before 5.6). For 5.6 and beyond, both DKMS and upstream kernels allow use of 15/16 of system memory. |
| | | IPC and RDMA capabilities are not yet enabled |
| | | Not tested by AMD to the same level as `rock-dkms` package |
| | | Does not include most up-to-date firmware |
# Disclaimer
AMD®, the AMD Arrow logo, AMD Instinct™, Radeon™, ROCm® and combinations thereof are trademarks of Advanced Micro Devices, Inc.
Linux® is the registered trademark of Linus Torvalds in the U.S. and other countries.
PCIe® is a registered trademark of PCI-SIG Corporation. Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.
Google® is a registered trademark of Google LLC.
Ubuntu and the Ubuntu logo are registered trademarks of Canonical Ltd.
Other product names used in this publication are for identification purposes only and may be trademarks of their respective companies.
# rocm-docs-redirects
Redirects ReadtheDocs Community documentation sites to Business
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.