Skip to content
GPUFlow Documentation

Linux AMD Setup

This guide covers Linux systems with AMD GPUs. AMD GPU support uses ROCm (Radeon Open Compute) for container access.

Prerequisites

Section titled “Prerequisites”
  • AMD RX 6000 series or newer GPU
  • Ubuntu 22.04+ or Fedora 39+
  • At least 16GB system RAM

Step 1: Check GPU compatibility

Section titled “Step 1: Check GPU compatibility”

Verify your AMD GPU is detected:

Terminal window
lspci | grep -i amd

Supported AMD GPUs for GPUFlow:

  • RX 6600, 6700, 6800, 6900 series
  • RX 7600, 7700, 7800, 7900 series
  • Radeon Pro W6000+ series

Step 2: Install AMD drivers

Section titled “Step 2: Install AMD drivers”

Install the AMD GPU drivers:

Terminal window
# Ubuntu/Debian
wget https://repo.radeon.com/amdgpu-install/22.40.5/ubuntu/jammy/amdgpu-install_5.4.50405-1_all.deb
sudo dpkg -i amdgpu-install_5.4.50405-1_all.deb
sudo apt update
sudo amdgpu-install --usecase=dkms,graphics,rocm


# Fedora
sudo dnf install https://repo.radeon.com/amdgpu-install/22.40.5/rhel/9.1/amdgpu-install-5.4.50405-1.el9.noarch.rpm
sudo amdgpu-install --usecase=dkms,graphics,rocm

Add your user to the render group:

Terminal window
sudo usermod -a -G render $USER
sudo usermod -a -G video $USER

Log out and back in for group changes to take effect.

Step 3: Verify ROCm installation

Section titled “Step 3: Verify ROCm installation”

Test that ROCm can see your GPU:

Terminal window
rocm-smi


# Should show your AMD GPU with temperature, usage, etc.

If rocm-smi is not found, install ROCm separately:

Terminal window
# Ubuntu
sudo apt install rocm-smi-lib


# Fedora
sudo dnf install rocm-smi

Step 4: Choose container runtime

Section titled “Step 4: Choose container runtime”

Select either Docker or Podman for running containers:

Docker + AMD

Standard Docker with ROCm container support

Continue with Docker →

Podman + AMD

Rootless containers with ROCm integration

Continue with Podman →

Known limitations

Section titled “Known limitations”

AMD GPU support has some current limitations:

  • Software compatibility: Fewer ML frameworks support ROCm compared to CUDA
  • Performance: May be slower than equivalent NVIDIA GPUs for AI workloads
  • Container ecosystem: Limited pre-built containers with ROCm support
  • Stability: ROCm drivers can be less stable than NVIDIA’s mature CUDA stack

What’s next

Section titled “What’s next”

After completing container runtime setup:

  1. Deploy GPUFlow provider with AMD GPU passthrough
  2. Create your provider account
  3. Configure GPU listings with AMD-specific notes