docs: document Docker Desktop /dev/kfd limitation on Linux

[nikhilsk]

Fixes #86

Problem

On Linux, when Docker Desktop is running, containers cannot access host GPU devices (/dev/kfd, /dev/dri). Users see:

docker: Error response from daemon: error gathering device information while adding custom device "/dev/kfd": no such file or directory

See issue #86.

Proposed workaround

• Use Docker Engine (e.g. docker.io or Docker's official repository) on Linux for GPU workloads so the daemon runs on the host and can expose /dev/kfd and /dev/dri.
• Or quit Docker Desktop (or log out and back in) so the host Docker Engine is used instead.

Test results

Verified on a Linux host with Docker Engine (default context, unix:///var/run/docker.sock):

1. Same command as in issue #86 (with --rm added so the container is removed on exit):

docker run -it --rm \
    --cap-add=SYS_PTRACE \
    --security-opt seccomp=unconfined \
    --device=/dev/kfd \
    --device=/dev/dri \
    --group-add video \
    --ipc=host \
    --shm-size 8G \
    rocm/pytorch:latest

• Container started successfully (no /dev/kfd error).
• Inside container: rocm-smi showed GPU devices; python -c "import torch; print(torch.cuda.is_available())" returned True.

2. Quick smoke test:

docker run --rm --device=/dev/kfd --device=/dev/dri rocm/rocm-terminal rocm-smi

• Image pulled, container ran, rocm-smi showed the GPU. No leftover container.

This confirms that when Docker Engine is used (not Docker Desktop), the documented workaround works and the exact run command from #86 succeeds.

[shiv-tyagi]

Thanks @nikhilsk. Posted comments. PTAL.

[shiv-tyagi]

LGTM. Thanks.