Switch GPU Util metric to DCGM_FI_PROF_GR_ENGINE_ACTIVE in NVIDIA DCGM Metrics Dashboard

[wabouhamad]

Is this a new feature, an improvement, or a change to existing functionality?

Improvement

Please provide a clear description of the problem this feature solves

The NVIDIA DCGM Metrics Dashboard on OpenShift 4.15 is using the DCGM_FI_DEV_GPU_UTIL metric which only shows 0% or 100% GPU Utilization. The more accurate metric is DCGM_FI_PROF_GR_ENGINE_ACTIVE. Need to switch metrics to report DCGM_FI_PROF_GR_ENGINE_ACTIVE for GPU utilization.

Feature Description

From a user prespective I need to see the more accurate GPU utilization when running a GPU workload and not 0% or 100% utilization. The current counter in the NVIDIA DCGM Dashboard on OpenShift is using an older metric DCGM_FI_PROF_GR_ENGINE_ACTIVE. Need to switch metrics to report DCGM_FI_PROF_GR_ENGINE_ACTIVE for GPU utilization.

Describe your ideal solution

Need to switch to report DCGM_FI_PROF_GR_ENGINE_ACTIVE metric for GPU utilization.

Additional context

No response