Folding@Home Client for Docker using AMD GPU

This project builds a Docker image for running the Folding@Home Client in headless mode as a Docker Container with access to the host system's AMD GPU. By design, this container does GPU only folding (no CPU folding).

Specifically this has been tested on my environment:

Arch Linux host with 5.6.7 and 5.9.2 kernels
AMD RX 580 GPU with open source amdgpu kernel driver on the host

Building the docker image locally

make

Optionally, you can confirm that an image was built locally with this:

docker image ls --filter reference=fah-amdgpu

Running

To run the container image as a Docker container to do some folding

Here we are bind mounting a config.xml file into the container.
Specifically not starting the container with the --rm flag since F@H work units can be resumed as they save state checkpoints to disk. So we want to be able to resume a container should we have to stop it for unrelated reasons (like a reboot).

docker run -d --device=/dev/dri -v `pwd`/config.xml:/home/fah/config.xml:ro --name fah fah-amdgpu

Monitoring

docker logs -f fah

Suspending folding

docker stop fah

Resuming folding

docker start fah

Purge/cleanup

Warning: F@H uses checkpoints to save progress on folding Work Units (WU) and credit for work is only granted at completion of the WU. Removing the container will wipe out checkpoints, so whatever progress was made on current WUs will be lost.

docker rm -f fah

Performance, Power, Fan Noise, and Heat

With GPU folding, there don't seem to be good F@H client options for throttling usage.

This is not a problem though, since in practice it's more elegant to just set a TDP (power) cap for your AMD GPU.

My RX 580 has a default power cap of 135 W. The power cap can be changed dynamically via sysfs without any special tools.

Example: To change the power cap to 50 W (50,000,000 μW), on the host (not in the container),

echo 50000000 | sudo tee /sys/class/drm/card0/device/hwmon/hwmon1/power1_cap

You may find it useful to install the rocm-smi tool on the host to check current power use.

Troubleshooting

To get an interactive shell to poke around:

docker run --rm -it --device=/dev/dri fah-amdgpu bash

Useful commands from the interactive container ...

List OpenCL devices
```
clinfo -l
```
Check Folding@Home Client configuration (confirming GPU count > 0)
```
FAHClient --info
```

Josholith/fah-amdgpu