Ollama with Vulkan Support for Intel Arc GPUs

Docker setup for running Ollama with Vulkan acceleration on Intel Arc GPUs (tested on Arc A770).

Features

• ✅ Latest Ollama with Vulkan support
• ✅ Full Intel Arc GPU acceleration
• ✅ Optional Open WebUI interface
• ✅ Easy configuration via environment variables
• ✅ Portainer compatible

Quick Start

Using Docker Compose

1. Clone this repository:

git clone https://github.com/lexiismadd/ollama-vulkan-arc.git
cd ollama-vulkan-arc

2. Copy and configure environment:

cp .env.example .env
# Edit .env with your preferences

3. Build and start:

docker compose build
docker compose up -d

4. Test:

docker exec ollama-vulkan-arc ollama pull llama3.2
docker exec ollama-vulkan-arc ollama run llama3.2

Using Portainer

See Portainer Setup below.

Configuration

Edit .env file to customize:

• OLLAMA_VERSION - Ollama version to build
• OLLAMA_NUM_GPU - GPU layer offloading (999 = all layers)
• OLLAMA_NUM_CTX - Context window size
• MEMORY_LIMIT - Maximum RAM usage
• ENABLE_WEBUI - Enable/disable Open WebUI

Portainer Setup

1. In Portainer, go to Stacks → Add stack
2. Choose Git Repository
3. Enter your repository URL: https://github.com/lexiismadd/ollama-vulkan-arc
4. Set Compose path: docker-compose.yml
5. Add environment variables or upload your .env file
6. Click Deploy the stack

See detailed Portainer instructions.

Monitoring GPU Usage

# Install tools
sudo apt install intel-gpu-tools

# Monitor GPU
sudo intel_gpu_top

Updating Ollama

1. Edit .env and update OLLAMA_VERSION
2. Rebuild:

docker compose down
docker compose build --no-cache
docker compose up -d

Troubleshooting

See TROUBLESHOOTING.md for common issues and solutions.

Requirements

• Intel Arc GPU (A770, A750, A380, etc.)
• Intel GPU drivers installed
• Docker or Podman
• Docker Compose

Performance

Expected tokens/s on Arc A770 16GB:

• Small models (3-8B): 50-100+ tokens/s
• Medium models (13-20B): 20-40 tokens/s
• Large models (30-70B): 5-15 tokens/s

Contributing

Feel free to open issues or submit pull requests!

License

MIT License - feel free to use and modify.

Acknowledgments

• Ollama
• Intel IPEX-LLM
• Vulkan support contributors