docker-compose.multi-gpu.yml

# Docker Compose for Multi-GPU Setups (Mixed Vendors)
#
# This configuration supports multiple GPUs from different vendors.
# SmarterRouter will automatically detect all GPUs and combine their VRAM.
#
# Supported combinations:
#   - NVIDIA + AMD
#   - NVIDIA + Intel
#   - AMD + Intel
#   - NVIDIA + AMD + Intel
#
# Note: Apple Silicon cannot be combined with other GPUs (runs on host only)
#
# Requirements:
#   - All relevant drivers installed for each GPU vendor
#   - NVIDIA Container Toolkit (for NVIDIA GPUs)
#   - ROCm runtime (for AMD GPUs)
#   - Intel i915 driver (for Intel GPUs)
#
# Quick Start:
#   1. Copy this file to your project root: cp docs/docker-compose.multi-gpu.yml docker-compose.yml
#   2. Copy environment template: cp ENV_DEFAULT .env
#   3. If using AMD GPUs, get group IDs and add to .env:
#        echo "RENDER_GID=$(getent group render | cut -d: -f3)" >> .env
#        echo "VIDEO_GID=$(getent group video | cut -d: -f3)" >> .env
#   4. Edit to uncomment only the GPU types you have
#   5. Run: docker-compose --compatibility up -d (for NVIDIA)
#      OR: docker-compose up -d (without NVIDIA)
#
# Verify GPU detection:
#   docker logs smarterrouter | grep -i "gpu\|vram"

services:
  smarterrouter:
    image: ghcr.io/peva3/smarterrouter:latest
    container_name: smarterrouter
    ports:
      - "11436:11436"
    env_file:
      - .env
    volumes:
      - ./data:/app/data:rw
      - type: tmpfs
        target: /tmp
    restart: unless-stopped
    networks:
      - smarterrouter-network
    healthcheck:
      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:11436/health')"]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 10s

    # =========================================================================
    # MULTI-GPU CONFIGURATION
    # Uncomment the sections for the GPUs you have
    # =========================================================================

    # -------------------------------------------------------------------------
    # NVIDIA GPUs (use with deploy.resources)
    # -------------------------------------------------------------------------
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all              # Use all NVIDIA GPUs
              capabilities: [gpu]

    # -------------------------------------------------------------------------
    # AMD GPUs (use with devices + group_add)
    # -------------------------------------------------------------------------
    # Uncomment if you have AMD GPUs:
    # devices:
    #   - /dev/kfd
    #   - /dev/dri
    # group_add:
    #   - ${RENDER_GID:-109}        # Get with: getent group render | cut -d: -f3
    #   - ${VIDEO_GID:-44}          # Get with: getent group video | cut -d: -f3

    # -------------------------------------------------------------------------
    # Intel GPUs (use with devices)
    # -------------------------------------------------------------------------
    # Uncomment if you have Intel Arc GPUs (add to devices list above):
    # devices:
    #   - /dev/dri

    # -------------------------------------------------------------------------
    # NVIDIA + AMD Combined
    # -------------------------------------------------------------------------
    # For systems with both NVIDIA and AMD GPUs, combine both configurations:
    # 
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: all
    #           capabilities: [gpu]
    # devices:
    #   - /dev/kfd
    #   - /dev/dri
    # group_add:                      # Required for AMD
    #   - ${RENDER_GID:-109}
    #   - ${VIDEO_GID:-44}

    # -------------------------------------------------------------------------
    # NVIDIA + Intel Combined
    # -------------------------------------------------------------------------
    # For systems with both NVIDIA and Intel GPUs:
    #
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: all
    #           capabilities: [gpu]
    # devices:
    #   - /dev/dri

    # -------------------------------------------------------------------------
    # AMD + Intel Combined
    # -------------------------------------------------------------------------
    # For systems with both AMD and Intel GPUs (no NVIDIA):
    #
    # devices:
    #   - /dev/kfd              # AMD
    #   - /dev/dri              # Both AMD and Intel
    # group_add:                # Required for AMD
    #   - ${RENDER_GID:-109}
    #   - ${VIDEO_GID:-44}

    # -------------------------------------------------------------------------
    # NVIDIA + AMD + Intel Combined
    # -------------------------------------------------------------------------
    # For systems with all three GPU types:
    #
    # deploy:
    #   resources:
    #     reservations:
    #       devices:
    #         - driver: nvidia
    #           count: all
    #           capabilities: [gpu]
    # devices:
    #   - /dev/kfd
    #   - /dev/dri
    # group_add:                      # Required for AMD
    #   - ${RENDER_GID:-109}
    #   - ${VIDEO_GID:-44}

    # =========================================================================
    # ENVIRONMENT CONFIGURATION
    # =========================================================================
    # environment:
    #   # Total VRAM limit across all GPUs (leave some buffer)
    #   - ROUTER_VRAM_MAX_TOTAL_GB=48
    #
    #   # Optional: Pin a model to keep in VRAM
    #   - ROUTER_PINNED_MODEL=phi3:mini
    #
    #   # Auto-unload when VRAM is pressured
    #   - ROUTER_VRAM_AUTO_UNLOAD_ENABLED=true
    #   - ROUTER_VRAM_UNLOAD_THRESHOLD_PCT=85

networks:
  smarterrouter-network:
    driver: bridge

# =========================================================================
# EXAMPLE CONFIGURATIONS
# =========================================================================

# Example: 2x NVIDIA RTX 3090 + 1x AMD Radeon Pro
# ---
# Total VRAM: 48GB (2x24) + 16GB = 64GB
# Recommended ROUTER_VRAM_MAX_TOTAL_GB: 60 (leave 4GB buffer)
#
# deploy:
#   resources:
#     reservations:
#       devices:
#         - driver: nvidia
#           count: all
#           capabilities: [gpu]
# devices:
#   - /dev/kfd
#   - /dev/dri
# group_add:
#   - ${RENDER_GID:-109}
#   - ${VIDEO_GID:-44}
# environment:
#   - ROUTER_VRAM_MAX_TOTAL_GB=60