How to Use Radeon RX 6000 GPUs to Run Ollama Models

By the end of this guide, your 6000-series GPU will run models like DeepSeek using Ollama with full GPU acceleration, bypassing unsupported architecture errors.

Requirements

• AMD GPU: RX 6000 series (e.g., 6700 XT with gfx1031 architecture)

• Linux OS: Ubuntu 22.04 (kernel 6.2+ recommended)

• ROCm 6.1: AMD’s compute framework for GPU acceleration

Step 1: Install ROCm on Ubuntu

1.1 Add the ROCm Repository

wget -q -O - https://repo.radeon.com/rocm/rocm.gpg.key | sudo gpg --dearmor -o /etc/apt/trusted.gpg.d/rocm.gpg  
echo 'deb [arch=amd64] https://repo.radeon.com/rocm/apt/6.1 jammy main' | sudo tee /etc/apt/sources.list.d/rocm.list

1.2 Install ROCm

sudo apt update && sudo apt install rocm

1.3 Grant GPU Access

sudo usermod -a -G render,video $USER  
sudo reboot

1.4 Confirm ROCm Detects Your GPU

/opt/rocm/bin/rocminfo | grep "Name: gfx1031"

Step 2: Force Ollama to Use Your GPU

2.1 Override Architecture Detection

Add this line to ~/.bashrc or ~/.zshrc:

export HSA_OVERRIDE_GFX_VERSION=10.3.0

Apply changes:

source ~/.bashrc

2.2 Launch Ollama with GPU Support

HSA_OVERRIDE_GFX_VERSION=10.3.0 ollama serve

Step 3: Run the DeepSeek Model

HSA_OVERRIDE_GFX_VERSION=10.3.0 ollama run deepseek-r1

Why This Works

• HSA_OVERRIDE_GFX_VERSION: Tricks ROCm into recognizing your GPU (gfx1031) as a supported architecture.

• ROCm 6.1: Provides critical compatibility updates for RDNA 2 GPUs.

Final Notes

This method bypasses Ollama’s default GPU blocklist, enabling near-native performance for RX 6000 GPUs. For updates, monitor Ollama’s GPU documentation.