Why is your GPU at 100% util but still slow? A zero-dependency Go CLI that diagnoses throttled, blocked, starved, or idle GPUs from DCGM + vLLM metrics — deterministic, replayable, and auditable.
kubernetes prometheus nvidia sre mfu gpu-monitoring finops gpu-utilization platform-engineering keda kv-cache inference-serving karpenter vllm llm-inference dcgm model-flops-utilization sm-occupancy gpu-fleet-management gpu-idle-detection
-
Updated
Sep 4, 2026 - Go