MLCommons / MLPerf Benchmarks

mlcommons.org

Home of the MLPerf benchmarks for training and inference speed. Results pages and methodology docs anchor notebooks comparing AI hardware and systems.

MLPerf is the closest thing the industry has to a shared benchmark for GPU and accelerator performance, since vendors submit results under common test rules instead of each picking its own conditions. A notebook comparing hardware for training or inference gets more defensible numbers here than from vendor marketing pages. Read the methodology docs alongside the results tables, since submission rules have changed across rounds and older comparisons don't always line up with newer ones.