In the latest MLCommons AI performance benchmark suite – MLPerf Inference v4.1 – Intel submitted results across six MLPerf benchmarks for 5th Gen Xeon Scalable processors and, for the first time, its Xeon 6 processors with Performance-cores (P-cores).
Intel Xeon 6 processors with P-cores achieved about 1.9x geomean performance improvement in AI performance compared with 5th Gen Xeon processors.
“The newest MLPerf results show how continued investment and resourcing is critical for improving AI performance,” says Pallavi Mahajan, Intel corporate vice-president and GM of Data Center and AI Software. “Over the past four years, we have raised the bar for AI performance on Intel Xeon processors by up to 17x based on MLPerf. As we near general availability later this year, we look forward to ramping Xeon 6 with our customers and partners.”
Why MLPerf results matter
AI systems require CPUs as a critical component to successfully deploy solutions across a variety of scenarios. Intel says Xeon provides a great solution for AI inference including classical machine learning and vector search embedding.
With MLPerf Inference v4.1, Intel submitted 5th Gen Intel Xeon processors and Xeon 6 processors with P-Cores on ResNet50, RetinaNet, 3DUNet, BERT, DLRM v2 and GPT-J. Compared with 5th Gen Intel Xeon, Xeon 6 provides an average of about 1.9x better AI inference performance across these six benchmarks. Intel says it continues to be the only server processor vendor to submit CPU results to MLPerf.
Over the past four years, Intel says it has made significant gains in AI performance with CPUs since it first submitted MLPerf results. Compared with 3rd Gen Xeon Scalable processors in 2021, Xeon 6 performs up to 17x better on natural language processing (BERT) and up to 15x better on computer vision (ResNet50) workloads.