View on GitHub ROCm_Logo

ROCm, a New Era in Open GPU Computing

Platform for GPU-Enabled HPC and Ultrascale Computing

ROCm Platform Installation Guide for Linux

Current ROCm Version: 2.0

Hardware Support

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.

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.

ROCm officially supports AMD GPUs that use following chips:

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:

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, by default, also require PCIe 3.0 with support for PCIe atomics, but they can operate in most cases without this capability.

At this time, the integrated GPUs in AMD APUs are not officially supported targets for ROCm. As descried below, “Carrizo”, “Bristol Ridge”, and “Raven Ridge” APUs are enabled in our upstream drivers and OpenCL runtime. However, they are not enabled in our HCC or HIP runtimes, 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.

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:

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 3.0 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):

kfd: skipped device 1002:7300, PCI rejects atomics

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
Not supported

The latest ROCm platform - ROCm 2.0

The latest supported version of the drivers, tools, libraries and source code for the ROCm platform have been released and are available from the following GitHub repositories:

Supported Operating Systems - New operating systems available

The ROCm 2.0.x platform supports the following operating systems:

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:

The upstream driver may be useful for running ROCm software on systems that are not compatible with the kernel driver available in AMD’s repositories. For users that have the option of using either AMD’s or the upstreamed driver, there are various tradeoffs to take into consideration:

  Using AMD’s rock-dkms package Using the upstream kernel driver
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 4.18 Limits GPU’s usage of system memory to 3/8 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

Installing from AMD ROCm repositories

AMD hosts both Debian and RPM repositories for the ROCm 2.0.x packages at this time.

The packages in the Debian repository have been signed to ensure package integrity.

ROCm Binary Package Structure

ROCm is a collection of software ranging from drivers and runtimes to libraries and developer tools. In AMD’s package distributions, these software projects are provided as a separate packages. This allows users to install only the packages they need, if they do not wish to install all of ROCm. These packages will install most of the ROCm software into /opt/rocm/ by default.

The packages for each of the major ROCm components are:

To make it easier to install ROCm, the AMD binary repos provide a number of meta-packages that will automatically install multiple other packages. For example, rocm-dkms is the primary meta-package that is used to install most of the base technology needed for ROCm to operate. It will install the rock-dkms kernel driver, and another meta-package (rocm-dev) which installs most of the user-land ROCm core components, support software, and development tools.

The rocm-utils meta-package will install useful utilities that, while not required for ROCm to operate, may still be beneficial to have. Finally, the rocm-libs meta-package will install some (but not all) of the libraries that are part of ROCm.

The chain of software installed by these meta-packages is illustrated below

rocm-dkms
 |-- rock-dkms
 \-- rocm-dev
      |--hsa-rocr-dev
      |--hsa-ext-rocr-dev
      |--rocm-device-libs
      |--rocm-utils
          |-- rocminfo
          |-- rocm-cmake
          \-- rocm-clang-ocl # This will cause OpenCL to be installed
      |--hcc
      |--hip_base
      |--hip_doc
      |--hip_hcc
      |--hip_samples
      |--rocm-smi
      |--hsakmt-roct
      |--hsakmt-roct-dev
      |--hsa-amd-aqlprofile
      |--comgr
      \--rocr_debug_agent

rocm-libs
 |-- rocblas
 |-- rocfft
 |-- rocrand
 \-- hipblas

These meta-packages are not required but may be useful to make it easier to install ROCm on most systems. Some users may want to skip certain packages. For instance, a user that wants to use the upstream kernel drivers (rather than those supplied by AMD) may want to skip the rocm-dkms and rock-dkms packages, and instead directly install rocm-dev.

Similarly, a user that only wants to install OpenCL support instead of HCC and HIP may want to skip the rocm-dkms and rocm-dev packages. Instead, they could directly install rock-dkms, rocm-opencl, and rocm-opencl-dev and their dependencies.

Ubuntu Support - installing from a Debian repository

The following directions show how to install ROCm on supported Debian-based systems such as Ubuntu 18.04. These directions may not work as written on unsupported Debian-based distributions. For example, newer versions of Ubuntu may not be compatible with the rock-dkms kernel driver. As such, users may want to skip the rocm-dkms and rock-dkms packages, as described above, and instead use the upstream kernel driver.

First make sure your system is up to date
sudo apt update
sudo apt dist-upgrade
sudo apt install libnuma-dev
sudo reboot
Add the ROCm apt repository

For Debian-based systems like Ubuntu, configure the Debian ROCm repository as follows:

wget -qO - http://repo.radeon.com/rocm/apt/debian/rocm.gpg.key | sudo apt-key add -
echo 'deb [arch=amd64] http://repo.radeon.com/rocm/apt/debian/ xenial main' | sudo tee /etc/apt/sources.list.d/rocm.list

The gpg key might change, so it may need to be updated when installing a new release. If the key signature verification is failed while update, please re-add the key from ROCm apt repository. The current rocm.gpg.key is not available in a standard key ring distribution, but has the following sha1sum hash:

f7f8147431c75e505c58a6f3a3548510869357a6 rocm.gpg.key

Install

Next, update the apt repository list and install the rocm-dkms meta-package:

sudo apt update
sudo apt install rocm-dkms
Next set your permissions

Users will need to be in the video group in order to have access to the GPU. As such, you should ensure that your user account is a member of the video group prior to using ROCm. You can find which groups you are a member of with the following command:

groups

To add yourself to the video group you will need the sudo password and can use the following command:

sudo usermod -a -G video $LOGNAME 

You may want to ensure that any future users you add to your system are put into the “video” group by default. To do that, you can run the following commands:

echo 'ADD_EXTRA_GROUPS=1' | sudo tee -a /etc/adduser.conf
echo 'EXTRA_GROUPS=video' | sudo tee -a /etc/adduser.conf

Once complete, reboot your system.

Test basic ROCm installation

After rebooting the system run the following commands to verify that the ROCm installation was successful. If you see your GPUs listed by both of these commands, you should be ready to go!

/opt/rocm/bin/rocminfo 
/opt/rocm/opencl/bin/x86_64/clinfo 

Note that, to make running ROCm programs easier, you may wish to put the ROCm binaries in your PATH.

echo 'export PATH=$PATH:/opt/rocm/bin:/opt/rocm/profiler/bin:/opt/rocm/opencl/bin/x86_64' | sudo tee -a /etc/profile.d/rocm.sh

If you have an install issue please read this FAQ.

Performing an OpenCL-only Installation of ROCm

Some users may want to install a subset of the full ROCm installation. In particular, if you are trying to install on a system with a limited amount of storage space, or which will only run a small collection of known applications, you may want to install only the packages that are required to run OpenCL applications. To do that, you can run the following installation command instead of the command to install rocm-dkms.

sudo apt-get install dkms rock-dkms rocm-opencl-dev
How to uninstall from Ubuntu 16.04 or Ubuntu 18.04

To uninstall the ROCm packages installed in the above directions, you can execute;

sudo apt autoremove rocm-dkms rocm-dev rocm-utils
Installing development packages for cross compilation

It is often useful to develop and test on different systems. For example, some development or build systems may not have an AMD GPU installed. In this scenario, you may prefer to avoid installing the ROCK kernel driver to your development system.

In this case, install the development subset of packages:

sudo apt update
sudo apt install rocm-dev

Note: To execute ROCm enabled apps you will require a system with the full ROCm driver stack installed

Using Debian-based ROCm with upstream kernel drivers

As described in the above section about upstream Linux kernel support, users may want to try installing ROCm user-level software without installing AMD’s custom ROCK kernel driver. Users who do want to use upstream kernels can run the following commands instead of installing rocm-dkms

sudo apt update
sudo apt install rocm-dev
echo 'SUBSYSTEM=="kfd", KERNEL=="kfd", TAG+="uaccess", GROUP="video"' | sudo tee /etc/udev/rules.d/70-kfd.rules

CentOS/RHEL 7 (7.4, 7.5, 7.6) Support

The following directions show how to install ROCm on supported RPM-based systems such as CentOS 7.6. These directions may not work as written on unsupported RPM-based distributions. For example, Fedora may work but may not be compatible with the rock-dkms kernel driver. As such, users may want to skip the rocm-dkms and rock-dkms packages, as described above, and instead use the upstream kernel driver.

Support for CentOS/RHEL 7 was added in ROCm 1.8, but ROCm requires a special runtime environment provided by the RHEL Software Collections and additional dkms support packages to properly install and run.

Preparing RHEL 7 (7.4, 7.5, 7.6) for installation

RHEL is a subscription-based operating system, and you must enable several external repositories to enable installation of the devtoolset-7 environment and the DKMS support files. These steps are not required for CentOS.

First, the subscription for RHEL must be enabled and attached to a pool id. Please see Obtaining an RHEL image and license page for instructions on registering your system with the RHEL subscription server and attaching to a pool id.

Second, enable the following repositories:

sudo subscription-manager repos --enable rhel-server-rhscl-7-rpms
sudo subscription-manager repos --enable rhel-7-server-optional-rpms
sudo subscription-manager repos --enable rhel-7-server-extras-rpms

Third, enable additional repositories by downloading and installing the epel-release-latest-7 repository RPM:

sudo rpm -ivh https://dl.fedoraproject.org/pub/epel/epel-release-latest-7.noarch.rpm
Install and setup Devtoolset-7

To setup the Devtoolset-7 environment, follow the instructions on this page:

https://www.softwarecollections.org/en/scls/rhscl/devtoolset-7/

Note that devtoolset-7 is a Software Collections package, and it is not supported by AMD.

Prepare CentOS/RHEL (7.4, 7.5, 7.6) for DKMS Install

Installing kernel drivers on CentOS/RHEL 7.4/7.5/7.6 requires dkms tool being installed:

sudo yum install -y epel-release
sudo yum install -y dkms kernel-headers-`uname -r` kernel-devel-`uname -r`
Installing ROCm on the system

It is recommended to remove previous ROCm installations before installing the latest version to ensure a smooth installation.

At this point ROCm can be installed on the target system. Create a /etc/yum.repos.d/rocm.repo file with the following contents:

[ROCm]
name=ROCm
baseurl=http://repo.radeon.com/rocm/yum/rpm
enabled=1
gpgcheck=0

The repo’s URL should point to the location of the repositories repodata database. Install ROCm components using these commands:

sudo yum install rocm-dkms

The rock-dkms component should be installed and the /dev/kfd device should be available on reboot.

Set up permissions

Ensure that your user account is a member of the “video” or “wheel” group prior to using the ROCm driver. You can find which groups you are a member of with the following command:

groups

To add yourself to the video (or wheel) group you will need the sudo password and can use the following command:

sudo usermod -a -G video $LOGNAME 

You may want to ensure that any future users you add to your system are put into the “video” group by default. To do that, you can run the following commands:

echo 'ADD_EXTRA_GROUPS=1' | sudo tee -a /etc/adduser.conf
echo 'EXTRA_GROUPS=video' | sudo tee -a /etc/adduser.conf

Current release supports CentOS/RHEL 7.4, 7.5, 7.6. If users want to update the OS version, they should completely remove ROCm packages before updating to the latest version of the OS, to avoid DKMS related issues.

Once complete, reboot your system.

Test basic ROCm installation

After rebooting the system run the following commands to verify that the ROCm installation was successful. If you see your GPUs listed by both of these commands, you should be ready to go!

/opt/rocm/bin/rocminfo
/opt/rocm/opencl/bin/x86_64/clinfo

Note that, to make running ROCm programs easier, you may wish to put the ROCm binaries in your PATH.

echo 'export PATH=$PATH:/opt/rocm/bin:/opt/rocm/profiler/bin:/opt/rocm/opencl/bin/x86_64' | sudo tee -a /etc/profile.d/rocm.sh

If you have an install issue please read this FAQ.

Performing an OpenCL-only Installation of ROCm

Some users may want to install a subset of the full ROCm installation. In particular, if you are trying to install on a system with a limited amount of storage space, or which will only run a small collection of known applications, you may want to install only the packages that are required to run OpenCL applications. To do that, you can run the following installation command instead of the command to install rocm-dkms.

sudo yum install rock-dkms rocm-opencl-devel
Compiling applications using HCC, HIP, and other ROCm software

To compile applications or samples, please use gcc-7.2 provided by the devtoolset-7 environment. To do this, compile all applications after running this command:

scl enable devtoolset-7 bash
How to uninstall ROCm from CentOS/RHEL 7.4, 7.5 and 7.6

To uninstall the ROCm packages installed by the above directions, you can execute:

sudo yum autoremove rocm-dkms rock-dkms
Installing development packages for cross compilation

It is often useful to develop and test on different systems. For example, some development or build systems may not have an AMD GPU installed. In this scenario, you may prefer to avoid installing the ROCK kernel driver to your development system.

In this case, install the development subset of packages:

sudo yum install rocm-dev

Note: To execute ROCm enabled apps you will require a system with the full ROCm driver stack installed

Using ROCm with upstream kernel drivers

As described in the above section about upstream Linux kernel support, use rs may want to try installing ROCm user-level software without installing AMD’s custom ROCK kernel driver. Users who do want to use upstream kernels can run the following commands instead of installing rocm-dkms

sudo yum install rocm-dev
echo 'SUBSYSTEM=="kfd", KERNEL=="kfd", TAG+="uaccess", GROUP="video"' | sudo tee /etc/udev/rules.d/70-kfd.rules

Known issues / workarounds

HCC: removed support for C++AMP in ROCm 2.0

HipCaffe is supported on single GPU configurations

The ROCm SMI library calls to rsmi_dev_power_cap_set() and rsmi_dev_power_profile_set() will not work for all but the first gpu in multi-gpu set ups.

Closed source components

The ROCm platform relies on a few closed source components to provide functionality such as HSA image support. These components are only available through the ROCm repositories, and they will either be deprecated or become open source components in the future. These components are made available in the following packages: