cost-model/protos/kubemodel/gpu.proto

syntax = "proto3";

import "kubemodel/diagnostic.proto";

package kubemodel;
option go_package = "github.com/opencost/opencost/core/pkg/model/pb/kubemodel";

// GPUDevice represents a GPU device with DCGM integration (provisioned resource)
// This tracks available GPU capacity on a node
message GPUDevice {
  // Identification
  string ID = 1;          // GPU UUID (hardware identifier)
  string nodeID = 2;      // Node hosting this GPU device

  // Properties
  int32 deviceNumber = 3;
  string modelName = 4;

  // GPU sharing information
  bool isShared = 6;
  float sharePercentage = 9;

  // Capacity metrics
  // GPU hours available
  float gpuHours = 10;

  // GPU request average percentage (0-100)
  float gpuRequestAverage = 11;

  // GPU usage average percentage (0-100)
  float gpuUsageAverage = 12;

  // GPU usage max percentage (0-100)
  float gpuUsageMax = 13;

  // GPU memory capacity in bytes
  int64 memoryBytes = 14;

  // Diagnostic information about this resource
  optional DiagnosticResult diagnostic = 99;
}

// GPUUsage represents GPU resources consumed by a container (allocated resource)
// This tracks actual GPU usage by containers for cost analysis
message GPUUsage {
  // Identification
  string containerID = 1;  // Container consuming GPU resources
  string gpuDeviceID = 2;  // Reference to the GPU device being used

  // Usage metrics
  // GPU usage in device-hours consumed
  float gpuHours = 3;

  // GPU request in percentage (0-100)
  float gpuRequestPercentage = 4;

  // GPU usage average percentage (0-100)
  float gpuUsageAverage = 5;

  // GPU usage max percentage (0-100)
  float gpuUsageMax = 6;

  // GPU memory usage in bytes
  int64 memoryBytesUsed = 7;

  // Diagnostic information about this resource
  optional DiagnosticResult diagnostic = 99;
}