Specifications for public cloud cluster acme-gcp-gke-prod hosting core pipelines and monitoring.
- System Pool: 3 nodes (e2-standard-4) dedicated to system workloads.
- Application Pool: 5 to 20 nodes (e2-highcpu-8) with autoscaling enabled.
- AI/ML Workload Pool: 2 nodes (g2-standard-8) with L4 GPUs for localized ONNX INT8 runtimes.