Add Google Cloud Platform as a GPU provider. Looking for contributors with GCP experience — I don't have access to GCP so can't test this myself.
What to implement
A new internal/providers/gcp/ package following the same pattern as internal/providers/aws/:
Discovery
- Compute Engine GPU VMs — A100, H100, L4, T4 instances (a2, a3, g2 machine families)
- Vertex AI endpoints — managed GPU inference endpoints
Metrics
- Cloud Monitoring — CPU utilization, GPU utilization (via
compute.googleapis.com/instance/gpu/utilization), network I/O
Pricing
- Add GCP GPU instance specs to
internal/pricing/ (on-demand + preemptible/spot prices)
How the codebase works
The existing analysis rules in internal/analysis/rules.go are provider-agnostic — they operate on models.GPUInstance. Once a GCP provider populates the same struct, all 9 rules work automatically.
Key files to reference:
internal/providers/aws/ec2.go — EC2 discovery pattern to followinternal/providers/aws/cloudwatch.go — metrics enrichment patterninternal/models/models.go — GPUInstance struct and Source type (add SourceGCE, SourceVertexAI)cmd/gpuaudit/main.go — CLI wiring (add --skip-gcp flag, call provider)
GCP SDK
Use cloud.google.com/go/compute and cloud.google.com/go/monitoring. Auth should use Application Default Credentials (gcloud auth application-default login).
Happy to review PRs and help with architecture decisions — just need someone who can test against real GCP GPU instances.
Add Google Cloud Platform as a GPU provider. Looking for contributors with GCP experience — I don't have access to GCP so can't test this myself.
What to implement
A new
internal/providers/gcp/package following the same pattern asinternal/providers/aws/:Discovery
Metrics
compute.googleapis.com/instance/gpu/utilization), network I/OPricing
internal/pricing/(on-demand + preemptible/spot prices)How the codebase works
The existing analysis rules in
internal/analysis/rules.goare provider-agnostic — they operate onmodels.GPUInstance. Once a GCP provider populates the same struct, all 9 rules work automatically.Key files to reference:
internal/providers/aws/ec2.go— EC2 discovery pattern to followinternal/providers/aws/cloudwatch.go— metrics enrichment patterninternal/models/models.go—GPUInstancestruct andSourcetype (addSourceGCE,SourceVertexAI)cmd/gpuaudit/main.go— CLI wiring (add--skip-gcpflag, call provider)GCP SDK
Use
cloud.google.com/go/computeandcloud.google.com/go/monitoring. Auth should use Application Default Credentials (gcloud auth application-default login).Happy to review PRs and help with architecture decisions — just need someone who can test against real GCP GPU instances.