Discussions

• 1 You must be logged in to vote

  🙏

  Helm: values structure for multi-GPU-model serving?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  How to measure energy per token, not just cost per token?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  Best way to pin vLLM version reproducibly in a container?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  Prometheus retention for GPU metrics — local vs remote-write?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  MIG vs full GPU for multiple small models?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  NVIDIA device plugin vs GPU Operator on k3s?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  Cold-start latency of scale-to-zero GPU pods — acceptable?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  How to label per-GPU cost in Grafana from DCGM?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  Kueue vs plain HPA for batching GPU jobs?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  DCGM: which metric best represents real GPU 'busy' time?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  FP8 vs FP16 quantization: real quality hit for a 7B?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  k3s vs Talos for a small GPU inference cluster?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  Terraform: cleanest way to teardown a RunPod pod on failure?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  vLLM: how much does --gpu-memory-utilization actually change throughput?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  k3s: how to expose DCGM exporter metrics to Prometheus cleanly?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1
• 1 You must be logged in to vote

  🙏

  How do you measure real €/token including idle GPU time?

  iacker asked Jul 10, 2026 in Q&A · Answered

  

  

  1