Show HN: Piqc – GPU waste scanner for LLM inference clusters
Category: infrastructure
Tags: gpu, kubernetes, cost-optimization, llm-inference, waste-scanner
Score: 7.5/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 8)
Piqc is a read-only GPU waste scanner for Kubernetes clusters running LLM inference, detecting idle allocations, tier misplacement, and dark capacity with dollar estimates. It combines intelligent inference deployment discovery with real-time GPU metrics to surface significant cost savings in under a minute. Its no-agents, one-command approach fills a clear gap in cost observability for AI infrastructure.
Target audience: devops, platform engineers, mlops
Repository: https://github.com/paralleliq/piqc · Python · NOASSERTION · 7 stars
View on Hacker News