MLPerf 3.1 includes major language model benchmarks for inference

MLCommons has expanded the MLPerf AI benchmarks suite with the addition of testing for large language models (LLMs) for inference and a new benchmark for assessing performance of storage systems for machine learning workloads. The organization, which offers a neutral platform for vendors to report on various aspects of AI performance, recently released the second major update of the MLPerf Inference 3.1 benchmarks. Many submitters have reportedly improved their performance by 20% or more over the previous 3.0 benchmark. Participating submitters include major tech companies such as ASUSTeK, Azure, Dell, Google, Intel, Lenovo, Nvidia, and more.