Just like with CUDA in docker containers or LXC, the host OS needs the proper drivers in order for the docker containers to function. 
 

unfortunately ROCm just isn’t as popular as CUDA so it makes finding info more difficult. 
 

I'm hoping to not have to use a VM for one or two apps when there are prebuilt docker images available with ROCm support. 

I am also seeking a solution to this. Trying to get a container to see the GPU has stumped me so far, and I haven't found any clear explanation on getting ROCm into Unraid/Slackware.

I know I’m bumping a really old post but I would have just ended up asking the same question if I started a new thread.

I am also wondering about this, I switched to AMD when I made my server (I wanted my main gaming rig back, but don't like to use CPU transcoding even for 1080p to 1080p). Made the switch and it's even more of a hassle than getting Nvidia products to "remove the spiked seat cushions". Anyways does anyone know anything new?

Did anyone found a solution here? I want use transcoding with my built in amd gpu but i guess without ROCm its not possible

Same topic here.

Wanted to use the integrated 680m in a Docker container (BOINC) but found no solution up to now

Install "Radeon Top" plugin from community apps to get the driver.

This plugin adds the tool 'radeontop' to your unRAID server and also enables your AMD GPU from the installation of this plugin on, so no editis to the 'go' file or creation of other files are necessary (please not that this plugin only enables the 'amdgpu' Kernel module and not the 'radeon' Kernel module).
To see the usage of your GPU open up the unRAID Terminal and type in 'radeontop' (without quotes).
This plugin satisfies installation prerequisites of the GPU Statistics plugin from Community Apps. With both plugins installed you can display AMD GPU utilization on the unRAID Dashboard.

Add two devices to your docker configuration to give it access: /dev/kfd and /dev/dri

https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/docker.html#accessing-gpus-in-containers

In order to grant access to GPUs from within a container, run your container with the following options:

docker run --device /dev/kfd --device /dev/dri --security-opt seccomp=unconfined <image>

The purpose of each option is as follows:

• --device /dev/kfd

  This is the main compute interface, shared by all GPUs.

• --device /dev/dri

  This directory contains the Direct Rendering Interface (DRI) for each GPU. To restrict access to specific GPUs, see Restricting GPU access.

• --security-opt seccomp=unconfined (optional)

  This option enables memory mapping, and is recommended for containers running in HPC environments.

  The performance of an application can vary depending on the assignment of GPUs and CPUs to the task. Typically, numactl is installed as part of many HPC applications to provide GPU/CPU mappings. This Docker runtime option supports memory mapping and can improve performance.

Install "Radeon Top" plugin from community apps to get the driver.

This plugin adds the tool 'radeontop' to your unRAID server and also enables your AMD GPU from the installation of this plugin on, so no editis to the 'go' file or creation of other files are necessary (please not that this plugin only enables the 'amdgpu' Kernel module and not the 'radeon' Kernel module).
To see the usage of your GPU open up the unRAID Terminal and type in 'radeontop' (without quotes).
This plugin satisfies installation prerequisites of the GPU Statistics plugin from Community Apps. With both plugins installed you can display AMD GPU utilization on the unRAID Dashboard.

Add two devices to your docker configuration to give it access: /dev/kfd and /dev/dri

https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/docker.html#accessing-gpus-in-containers

In order to grant access to GPUs from within a container, run your container with the following options:

docker run --device /dev/kfd --device /dev/dri --security-opt seccomp=unconfined <image>

The purpose of each option is as follows:

• --device /dev/kfd

  This is the main compute interface, shared by all GPUs.

• --device /dev/dri

  This directory contains the Direct Rendering Interface (DRI) for each GPU. To restrict access to specific GPUs, see Restricting GPU access.

• --security-opt seccomp=unconfined (optional)

  This option enables memory mapping, and is recommended for containers running in HPC environments.

  The performance of an application can vary depending on the assignment of GPUs and CPUs to the task. Typically, numactl is installed as part of many HPC applications to provide GPU/CPU mappings. This Docker runtime option supports memory mapping and can improve performance.

What version of rocm does radeon top plugin provide? Anyone have success? I ahve a mi50 on the way and would like unraid to have the necessary rocm and or amdgpu vulkan.

On 7/8/2025 at 6:44 AM, Unoid said:

What version of rocm does radeon top plugin provide? Anyone have success? I ahve a mi50 on the way and would like unraid to have the necessary rocm and or amdgpu vulkan.

no idea but Ollama:rocm doesn’t work with my 780m. It works for transcodding but ollama doesn’t see the device.

On 5/29/2025 at 10:26 AM, lbenedetto said:

Install "Radeon Top" plugin from community apps to get the driver.

Add two devices to your docker configuration to give it access: /dev/kfd and /dev/dri

https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/docker.html#accessing-gpus-in-containers

Sadly this doesn't work. Ollama still doesn't see the gpu.

AMD Ryzen AI 9 HX PRO 370 w/ Radeon 890M @ 2000 MHz
+
Radeon Top Plugin

In particular try adding one or both of these environment variables....
-e 'HSA_OVERRIDE_GFX_VERSION'='11.0.0'
-e 'ROCR_VISIBLE_DEVICES'='0'

As well as passing through the /dev/kfd and /dev/dri devices.  

docker run
  -d
  --name='ollama'
  --net='bridge'
  --pids-limit 2048
  -e TZ="America/New_York"
  -e HOST_OS="Unraid"
  -e HOST_HOSTNAME="UNRAID"
  -e HOST_CONTAINERNAME="ollama"
  -e 'OLLAMA_ORIGINS'='*'
  -e 'HSA_OVERRIDE_GFX_VERSION'='11.0.0'
  -e 'ROCR_VISIBLE_DEVICES'='0'
  -p '11434:11434/tcp'
  -v '/mnt/user/appdata/ollama':'/root/.ollama':'rw'
  --device='/dev/kfd'
  --device='/dev/dri' 'ollama/ollama:rocm'

ime=2025-12-10T04:00:45.107Z level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=ROCm compute=gfx1100 name=ROCm0 description="AMD Radeon Graphics" libdirs=ollama,rocm driver=60342.13 pci_id=0000:cd:00.0 type=iGPU total="8.0 GiB" available="7.9 GiB"

time=2025-12-10T04:00:45.107Z level=INFO source=routes.go:1638 msg="entering low vram mode" "total vram"="8.0 GiB" threshold="20.0 GiB"

Doesnt' seem to want to dynamically assign more memory on demand... going to go into the bios and assign more memory do more testing. But def seems to be using the GPU.

8 hours ago, DaCeige said:

AMD Ryzen AI 9 HX PRO 370 w/ Radeon 890M @ 2000 MHz
+
Radeon Top Plugin

In particular try adding one or both of these environment variables....
-e 'HSA_OVERRIDE_GFX_VERSION'='11.0.0'
-e 'ROCR_VISIBLE_DEVICES'='0'

As well as passing through the /dev/kfd and /dev/dri devices.  

docker run
  -d
  --name='ollama'
  --net='bridge'
  --pids-limit 2048
  -e TZ="America/New_York"
  -e HOST_OS="Unraid"
  -e HOST_HOSTNAME="UNRAID"
  -e HOST_CONTAINERNAME="ollama"
  -e 'OLLAMA_ORIGINS'='*'
  -e 'HSA_OVERRIDE_GFX_VERSION'='11.0.0'
  -e 'ROCR_VISIBLE_DEVICES'='0'
  -p '11434:11434/tcp'
  -v '/mnt/user/appdata/ollama':'/root/.ollama':'rw'
  --device='/dev/kfd'
  --device='/dev/dri' 'ollama/ollama:rocm'

ime=2025-12-10T04:00:45.107Z level=INFO source=types.go:42 msg="inference compute" id=0 filter_id=0 library=ROCm compute=gfx1100 name=ROCm0 description="AMD Radeon Graphics" libdirs=ollama,rocm driver=60342.13 pci_id=0000:cd:00.0 type=iGPU total="8.0 GiB" available="7.9 GiB"

time=2025-12-10T04:00:45.107Z level=INFO source=routes.go:1638 msg="entering low vram mode" "total vram"="8.0 GiB" threshold="20.0 GiB"

Doesnt' seem to want to dynamically assign more memory on demand... going to go into the bios and assign more memory do more testing. But def seems to be using the GPU.

I have the same setup and used the same environment variables....

however, I allocated 16GB of VRAM.. but when I run load a model in Ollama, I noticed that the VRAM doesnt increased. Don't seem to be normal?