See the Model FLOPs Utilization gap behind your GPU's "100% utilization": the GPU spend nvidia-smi and DCGM hide. No root access required.
gpu cuda prometheus nvidia observability mfu gpu-monitoring finops gpu-utilization llm llm-serving vllm llm-inference sglang dcgm model-flops-utilization gpu-efficiency
-
Updated
Jun 14, 2026 - Go