Home > Servers > PowerEdge Components > White Papers > Developing and Deploying Vision AI with Dell and NVIDIA Metropolis > Overview
The following figure shows the utilization in terms of GPU compute, GPU memory, and encoder and decoder processing for tests A through E. For tests A to D, there is capacity left in in GPU memory, and encoder and decoder processing, even though the compute limit was reached. For Test E, we set up intervals from 0 to 1, to skip one frame for inference. We observed that with this configuration, the decoder processing, along with the GPU compute, both become limiting factors on inference throughput.
The following figures show the plots generated with the NVDIA nvtop profiling tool.