modelgrid/ts/interfaces/gpu.ts

/**
 * ModelGrid GPU Interfaces
 *
 * Defines types for GPU detection and management.
 */

/**
 * GPU vendor type
 */
export type TGpuVendor = 'nvidia' | 'amd' | 'intel' | 'unknown';

/**
 * GPU information detected from the system
 */
export interface IGpuInfo {
  /** Unique identifier for this GPU */
  id: string;
  /** GPU vendor */
  vendor: TGpuVendor;
  /** GPU model name (e.g., "NVIDIA GeForce RTX 4090") */
  model: string;
  /** Total VRAM in MB */
  vram: number;
  /** Driver version (if available) */
  driverVersion?: string;
  /** CUDA version (NVIDIA only) */
  cudaVersion?: string;
  /** Compute capability (NVIDIA only, e.g., "8.9") */
  computeCapability?: string;
  /** ROCm version (AMD only) */
  rocmVersion?: string;
  /** oneAPI version (Intel only) */
  oneApiVersion?: string;
  /** PCI slot identifier */
  pciSlot: string;
  /** PCI bus ID (e.g., "0000:01:00.0") */
  pciBusId?: string;
  /** GPU index in the system */
  index: number;
}

/**
 * Real-time GPU status
 */
export interface IGpuStatus {
  /** GPU identifier */
  id: string;
  /** Current GPU utilization percentage (0-100) */
  utilization: number;
  /** Current memory usage in MB */
  memoryUsed: number;
  /** Total memory in MB */
  memoryTotal: number;
  /** Memory usage percentage */
  memoryPercent: number;
  /** Current temperature in Celsius */
  temperature: number;
  /** Current power usage in Watts */
  powerUsage: number;
  /** Power limit in Watts */
  powerLimit: number;
  /** Fan speed percentage (if available) */
  fanSpeed?: number;
  /** GPU clock speed in MHz */
  gpuClock?: number;
  /** Memory clock speed in MHz */
  memoryClock?: number;
  /** Last update timestamp */
  lastUpdate: number;
}

/**
 * Combined GPU information and status
 */
export interface IGpuFullStatus extends IGpuInfo {
  /** Real-time status */
  status: IGpuStatus;
  /** Container ID assigned to this GPU (if any) */
  assignedContainer?: string;
  /** Whether the GPU is available for use */
  available: boolean;
  /** Health status */
  health: 'healthy' | 'warning' | 'error' | 'unknown';
  /** Health message (if warning or error) */
  healthMessage?: string;
}

/**
 * System information including all GPUs
 */
export interface ISystemInfo {
  /** System hostname */
  hostname: string;
  /** CPU model name */
  cpuModel: string;
  /** Number of CPU cores */
  cpuCores: number;
  /** Total RAM in MB */
  ramTotal: number;
  /** Available RAM in MB */
  ramAvailable: number;
  /** Operating system */
  os: string;
  /** Kernel version */
  kernelVersion: string;
  /** List of detected GPUs */
  gpus: IGpuInfo[];
  /** Docker version (if installed) */
  dockerVersion?: string;
  /** NVIDIA Container Toolkit version (if installed) */
  nvidiaContainerVersion?: string;
}

/**
 * Driver status for a vendor
 */
export interface IDriverStatus {
  /** GPU vendor */
  vendor: TGpuVendor;
  /** Whether the driver is installed */
  installed: boolean;
  /** Driver version (if installed) */
  version?: string;
  /** CUDA/ROCm/oneAPI toolkit version (if installed) */
  toolkitVersion?: string;
  /** Container runtime support (e.g., nvidia-docker) */
  containerSupport: boolean;
  /** Container runtime version */
  containerRuntimeVersion?: string;
  /** List of detected issues */
  issues: string[];
}