MLCommons Releases New MLPerf Storage v3.0 Benchmark Results

▤ Summary

MLCommons released MLPerf Storage v3.0 benchmark results for AI storage systems. The new version adds tests for KV cache and vector database workloads and introduces S3 object storage support alongside POSIX. Nineteen organizations submitted results, including eleven first-time participants such as Azure, Nebius, NVIDIA, and XSKY. The benchmark highlights storage performance, power efficiency, and procurement guidance for teams provisioning AI training and inference infrastructure.

▦ Linked Companies