Ray on ROCm installation#
2026-10-29
4 min read time
System requirements#
To use Ray 2.59.0, you need the following prerequisites:
Install Ray#
To install Ray on ROCm, you have the following options:
Build your own Docker image#
Clone the AMD-Ecosystem/ray repository:
git clone https://github.com/AMD-Ecosystem/ray.git -b release/2.59.0.amd0
Build the Docker container using the Dockerfile in the
ray/dockerdirectory:cd ray docker build -f docker/Dockerfile.rocm -t my-rocm-ray .
Launch and connect to the container:
docker run --rm -it \ --device=/dev/dri \ --device=/dev/kfd \ -p 8265:8265 \ --group-add video \ --cap-add=SYS_PTRACE \ --security-opt seccomp=unconfined \ --privileged \ -v $HOME/.ssh:/root/.ssh \ -v $HOME:$HOME \ --shm-size 128G \ -w $PWD \ --name rocm_verl \ my-rocm-ray \ /bin/bash
Note
The
--shm-sizeparameter allocates shared memory for the container. It can be adjusted based on your system’s resources.Warning
--privileged,--cap-add=SYS_PTRACE, and--security-opt seccomp=unconfinedreduce container isolation. The command also mounts$HOMEand$HOME/.sshinto the container. Use this configuration on a trusted development host. On a shared or production host, drop the flags that your workload does not require.Verify the installed Ray version:
pip3 freeze | grep ray
Expected output:
ray @ https://s3-us-west-2.amazonaws.com/ray-wheels/master/c50286910448ba6005a85da1e22eca94bba38428/ray-3.0.0.dev0-cp314-cp314-manylinux2014_x86_64.whl#sha256=d6ae2b51cb649f0c4869cbd7d8688380fb26c61158d9b997e5657dfe97ae7b48
Install Ray on bare metal or a custom container#
Follow these steps if you prefer to install ROCm manually on your host system or in a custom container.
Install ROCm. Follow the ROCm installation guide to install ROCm on your system.
Once installed, verify your ROCm installation using:
amd-smi
Expected output:
+------------------------------------------------------------------------------+ | AMD-SMI 27.1.0+7da43026 | | amdgpu Version: 6.19.14.31400100 | | ROCm Version: 10.1.0 | | VBIOS Version: 00160474 | | FW PLDM: 01.25.04.02 | | Platform: Linux Baremetal | |-------------------------------------+----------------------------------------| | BDF GPU-Name | Mem-Uti Temp UEC Power-Usage | | GPU HIP-ID OAM-ID Partition-Mode | GFX-Uti Fan Mem-Usage | |=====================================+========================================| | 0000:09:00.0 AMD Instinct MI325X | 0 % 54 °C 0 139/1000 W | | 0 1 3 SPX/NPS1 | 0 % N/A 283/262128 MB | |-------------------------------------+----------------------------------------| | 0000:1b:00.0 AMD Instinct MI325X | 0 % 64 °C 0 163/1000 W | | 1 3 2 SPX/NPS1 | 0 % N/A 284/262128 MB | |-------------------------------------+----------------------------------------| | 0000:6b:00.0 AMD Instinct MI325X | 0 % 63 °C 0 171/1000 W | | 2 2 1 SPX/NPS1 | 0 % N/A 283/262128 MB | |-------------------------------------+----------------------------------------| | 0000:79:00.0 AMD Instinct MI325X | 0 % 51 °C 0 142/1000 W | | 3 0 0 SPX/NPS1 | 0 % N/A 283/262128 MB | |-------------------------------------+----------------------------------------| | 0000:89:00.0 AMD Instinct MI325X | 0 % 53 °C 0 138/1000 W | | 4 5 7 SPX/NPS1 | 0 % N/A 283/262128 MB | |-------------------------------------+----------------------------------------| | 0000:9a:00.0 AMD Instinct MI325X | 0 % 62 °C 0 167/1000 W | | 5 7 6 SPX/NPS1 | 0 % N/A 283/262128 MB | |-------------------------------------+----------------------------------------| | 0000:ea:00.0 AMD Instinct MI325X | 0 % 65 °C 0 173/1000 W | | 6 6 5 SPX/NPS1 | 0 % N/A 283/262128 MB | |-------------------------------------+----------------------------------------| | 0000:f9:00.0 AMD Instinct MI325X | 0 % 55 °C 0 151/1000 W | | 7 4 4 SPX/NPS1 | 0 % N/A 283/262128 MB | +-------------------------------------+----------------------------------------+ +------------------------------------------------------------------------------+ | Processes: | | GPU PID Process Name GTT_MEM VRAM_MEM MEM_USAGE CU % SDMA | |==============================================================================| | No running processes found | +------------------------------------------------------------------------------+
Install the required version of Ray with ROCm support using pip:
pip install -U \ "ray[default,serve] @ https://s3-us-west-2.amazonaws.com/ray-wheels/master/c50286910448ba6005a85da1e22eca94bba38428/ray-3.0.0.dev0-cp314-cp314-manylinux2014_x86_64.whl"
Verify the installed Ray version:
pip3 freeze | grep ray
Expected output:
ray @ https://s3-us-west-2.amazonaws.com/ray-wheels/master/c50286910448ba6005a85da1e22eca94bba38428/ray-3.0.0.dev0-cp314-cp314-manylinux2014_x86_64.whl#sha256=d6ae2b51cb649f0c4869cbd7d8688380fb26c61158d9b997e5657dfe97ae7b48
Build Ray from source#
Follow the Building Ray from source guide to build Ray with ROCm support from source.
Test the Ray installation#
Ray unit tests are optional for validating your installation if you used a prebuilt Docker image from AMD ROCm Docker Hub. To run unit tests manually and validate your installation fully, follow these steps:
After launching the container, test whether Ray detects ROCm devices as expected.
python3 -c "import ray; ray.init(); print(ray.cluster_resources())"
If the setup is successful, the output should list all available ROCm devices.
Expected output (for example, on the MI325X node):
{'object_store_memory': 20401094656.0, 'memory': 2411852390400.0, 'CPU': 384.0, 'node:10.194.129.123': 1.0, 'GPU': 8.0, 'accelerator_type:AMD-Instinct-MI325X-OAM': 1.0, 'node:__internal_head__': 1.0}