GPU Configuration Guide

This NixOS template includes comprehensive GPU support for AMD, NVIDIA, and Intel graphics cards, with different optimization profiles for desktop/gaming vs AI/compute workloads.

Overview

The GPU configuration system provides optimized settings for different use cases. You must manually enable the appropriate GPU modules:

Desktop/Gaming: Optimized for graphics, gaming, and multimedia
AI/Compute: Optimized for machine learning, CUDA/ROCm, and compute workloads
Server: Headless compute optimizations for AI/ML servers

Quick Start

Desktop Configuration

For desktop systems, edit hosts/your-hostname/configuration.nix:

modules.hardware.gpu = {
  # Optional: Enable detection service for logging GPU info
  autoDetect = true;
  profile = "desktop";  # or "gaming" for gaming-focused optimizations

  # You must manually enable your GPU type:
  # nvidia.enable = true;  # For NVIDIA GPUs
  # amd.enable = true;     # For AMD GPUs
  # intel.enable = true;   # For Intel GPUs
};

Server Configuration

For AI/compute servers, edit hosts/your-hostname/configuration.nix:

modules.hardware.gpu = {
  # Optional: Enable detection service for logging GPU info
  autoDetect = true;
  profile = "ai-compute";  # or "server-compute"

  # Example NVIDIA AI server configuration:
  nvidia = {
    enable = true;
    compute = {
      cuda = true;
      cudnn = true;
      containers = true;  # Docker/Podman NVIDIA support
    };
  };
};

Manual Configuration

AMD GPU Configuration

For AMD graphics cards (Radeon RX, Vega, RDNA series):

modules.hardware.gpu = {
  profile = "desktop";  # or "ai-compute"

  amd = {
    enable = true;
    model = "auto";  # auto, rdna3, rdna2, rdna1, vega, polaris

    # Gaming/Desktop features
    gaming = {
      enable = true;
      vulkan = true;      # Vulkan API support
      opengl = true;      # OpenGL optimizations
    };

    # AI/Compute features (for server profile)
    compute = {
      enable = true;      # Enable for AI workloads
      rocm = true;        # ROCm platform
      openCL = true;      # OpenCL support
      hip = true;         # HIP runtime
    };

    powerManagement = {
      enable = true;
      profile = "auto";   # auto, low, high, manual
    };
  };
};

AMD ROCm Support: AMD’s ROCm platform provides CUDA-like functionality for AI/ML workloads on AMD GPUs.

NVIDIA GPU Configuration

For NVIDIA graphics cards (GeForce, RTX, Quadro, Tesla):

modules.hardware.gpu = {
  profile = "desktop";  # or "ai-compute"

  nvidia = {
    enable = true;
    driver = "stable";    # stable, beta, production, open, legacy_470

    hardware = {
      model = "auto";     # auto, rtx40, rtx30, rtx20, gtx16, gtx10
      powerLimit = null;  # Set power limit in watts (e.g., 300)
    };

    # Gaming/Desktop features
    gaming = {
      enable = true;
      gsync = true;       # G-SYNC support
      rtx = true;         # RTX features (ray tracing, DLSS)
      nvenc = true;       # Hardware video encoding

      # Laptop hybrid graphics (PRIME)
      prime = {
        enable = true;    # Enable for laptops with Intel + NVIDIA
        offload = true;   # Offload mode (battery friendly)
        sync = false;     # Sync mode (always use NVIDIA)
      };
    };

    # AI/Compute features
    compute = {
      enable = true;      # Enable for AI workloads
      cuda = true;        # CUDA toolkit
      cudnn = true;       # cuDNN deep learning library
      tensorrt = true;    # TensorRT inference optimization
      containers = true;  # NVIDIA container runtime
      mig = false;        # Multi-Instance GPU (A100, H100)
    };

    # Professional features
    professional = {
      enable = true;      # Professional optimizations
      nvv4l2 = true;      # Video4Linux2 support
    };
  };
};

NVIDIA Container Support: Automatically configures Docker/Podman to use NVIDIA GPUs in containers.

Intel GPU Configuration

For Intel integrated graphics (UHD, Iris, Arc):

modules.hardware.gpu = {
  profile = "desktop";

  intel = {
    enable = true;
    generation = "auto";  # auto, arc, xe, iris-xe, iris-plus, uhd, hd

    # Desktop features
    desktop = {
      enable = true;
      vaapi = true;       # Hardware video acceleration
      vulkan = true;      # Vulkan API support
      opengl = true;      # OpenGL optimizations
    };

    # Compute features (Intel Arc/Xe)
    compute = {
      enable = true;      # For Intel Arc GPUs
      oneapi = true;      # Intel OneAPI toolkit
      opencl = true;      # OpenCL support
      level_zero = true;  # Level Zero API
    };

    powerManagement = {
      enable = true;
      rc6 = true;         # RC6 power states
      fbc = true;         # Frame Buffer Compression
      psr = true;         # Panel Self Refresh
    };
  };
};

Intel Arc Support: Intel’s discrete Arc GPUs support AI/compute workloads through OneAPI.

Multi-GPU Configuration

For systems with multiple GPUs:

modules.hardware.gpu = {
  profile = "gaming";  # or "ai-compute"

  # Enable multiple GPU types
  amd.enable = true;
  nvidia.enable = true;

  # Multi-GPU configuration
  multiGpu = {
    enable = true;
    primary = "nvidia";  # Which GPU handles display: amd, nvidia, intel
  };
};

Profile Optimizations

Desktop Profile

Gaming optimizations (high performance, low latency)
Multimedia acceleration (video decode/encode)
Display optimizations (G-SYNC, FreeSync)
User-friendly tools and monitoring

Gaming Profile

Maximum performance settings
Gaming-specific optimizations (GameMode, MangoHud)
Low-latency configurations
Overclocking support

AI-Compute Profile

CUDA/ROCm/OneAPI support
Machine learning libraries
Container runtime support
Compute-optimized power management
Development tools and profilers

Server-Compute Profile

Headless optimizations
Maximum compute throughput
Container and orchestration support
Monitoring and management tools
Power efficiency optimizations

Common Use Cases

Gaming Desktop with NVIDIA RTX

modules.hardware.gpu = {
  profile = "gaming";
  nvidia = {
    enable = true;
    gaming = {
      enable = true;
      gsync = true;
      rtx = true;
    };
  };
};

AI Development Server with AMD GPU

modules.hardware.gpu = {
  profile = "ai-compute";
  amd = {
    enable = true;
    compute = {
      enable = true;
      rocm = true;
      hip = true;
    };
  };
};

Laptop with Intel + NVIDIA Hybrid Graphics

modules.hardware.gpu = {
  profile = "desktop";
  intel.enable = true;
  nvidia = {
    enable = true;
    gaming.prime = {
      enable = true;
      offload = true;  # Battery-friendly
    };
  };
  multiGpu = {
    enable = true;
    primary = "intel";  # Intel for display, NVIDIA for compute
  };
};

Multi-GPU AI Server

modules.hardware.gpu = {
  profile = "server-compute";
  nvidia = {
    enable = true;
    compute = {
      enable = true;
      cuda = true;
      cudnn = true;
      containers = true;
      mig = true;  # For A100/H100 GPUs
    };
    hardware.powerLimit = 400;  # Power limit per GPU
  };
  multiGpu.enable = true;
};

Hardware Detection

The system automatically detects GPUs using:

PCI device enumeration
Driver availability checks
Hardware capability detection

Detection results are logged and available at /run/gpu-info/detected.

Troubleshooting

Check GPU Detection

# View detected GPUs
cat /run/gpu-info/detected

# Check PCI devices
lspci | grep -i vga
lspci | grep -i 3d

# Check loaded drivers
lsmod | grep -E "(nvidia|amdgpu|i915)"

NVIDIA Issues

# Check NVIDIA driver
nvidia-smi

# Check CUDA
nvcc --version

# Container support
docker run --rm --gpus all nvidia/cuda:11.0-base-ubuntu20.04 nvidia-smi

AMD Issues

# Check AMD GPU
radeontop -d -

# Check ROCm
rocm-smi

# Check OpenCL
clinfo

Intel Issues

# Check Intel GPU
intel_gpu_top

# Check VA-API
vainfo

Performance Tuning

Gaming Optimizations

G-SYNC/FreeSync enabled automatically
Game mode optimizations
Shader caching configured
Low-latency settings

AI/Compute Optimizations

Memory management tuned for large datasets
Compute-focused power profiles
Container runtime optimizations
Development tool integration

Power Management

Automatic GPU power states
Temperature monitoring
Configurable power limits
Thermal throttling protection

Container Support

NVIDIA Container Runtime is automatically configured for:

Docker with --gpus all flag
Podman with NVIDIA support
Kubernetes GPU scheduling
AI/ML container workflows

Example usage:

# Docker with GPU
docker run --rm --gpus all tensorflow/tensorflow:latest-gpu python -c "import tensorflow as tf; print(tf.config.list_physical_devices('GPU'))"

# Podman with GPU
podman run --rm --device nvidia.com/gpu=all pytorch/pytorch:latest python -c "import torch; print(torch.cuda.is_available())"

Monitoring Tools

Automatically installed based on GPU type:

nvtop: Universal GPU monitoring (NVIDIA, AMD, Intel)
nvidia-smi: NVIDIA management
radeontop: AMD monitoring
intel_gpu_top: Intel monitoring
MangoHud: Gaming overlay

Getting Help

Check system logs: journalctl -u gpu-detection
Verify hardware detection: lspci | grep -i gpu
Test GPU functionality with appropriate tools
Review module configuration in your host config
Check NixOS hardware database for your specific GPU