This paper describes the results of our performance testing of Dell EMC Ready Stack for Red Hat OpenShift Container Platform 4.3, accelerated with NVIDIA GPUs. This configuration provides an enterprise-grade container orchestration platform with optimized performance for AI and machine learning training and inference workloads.
We validated the reference architecture by running AI workloads MLPerf Training v0.6 and Inference v0.5 and comparing the performance of our reference architecture to NVIDIA MLPerf Training and Dell MLPerf Inference published results, which do not include a container orchestration layer. Our MLPerf Training and Inference benchmark results achieved performance comparable to the NVIDIA published MLPerf results on a per-GPU basis. This outcome demonstrates that OpenShift Container Platform software adds valuable benefits for running AI and machine learning workloads in production environments.