zeekay dde006617a Initial commit: EKS GPU CI infrastructure ...

EKS cluster with Karpenter v1 NodePool that scales g4dn.xlarge spot
instances 0..1 on demand for CUDA workflows. ARC RunnerScaleSet
registers as [self-hosted, linux, amd64, cuda, gpu, t4].

Components:
- terraform/ — VPC, EKS 1.31, system node group, Karpenter IAM/SQS
- helm/karpenter-nodepool.yaml — v1 NodePool + EC2NodeClass for GPU spot
- helm/nvidia-device-plugin-values.yaml — DS with GPU taint toleration
- helm/arc-values.yaml + helm/runner-scaleset.yaml — ARC v0.9.x
- bootstrap.sh — one-shot deploy
- Makefile — init/apply/destroy/verify/cost-estimate

Cost: ~$138/mo idle (EKS control plane + tiny system node + NAT),
~$0.027/build for a 10-minute g4dn.xlarge spot run.

2026-02-27 09:00:00 -08:00

..

main.tf

Initial commit: EKS GPU CI infrastructure

2026-02-27 09:00:00 -08:00

variables.tf

Initial commit: EKS GPU CI infrastructure

2026-02-27 09:00:00 -08:00

versions.tf

Initial commit: EKS GPU CI infrastructure

2026-02-27 09:00:00 -08:00