MLCommons releases MLPerf Storage v3.0 benchmark results
4 companies named across 8 categories, one of 11 articles referencing MLCommons. Previous coverage: MLCommons Releases MLPerf Training v6.0 Results (Jun 2026).
Best suited for
- Job function
- Chief Data / Analytics / AI Officer
- Seniority
- Director
- Persona
- Data / AI Operations Leader
- Buyer role
- Decision Maker / Budget Holder
- Buyer journey
- Need to Buy
- Adoption curve
- Early Majority
- Technology maturity
- Operational Expansion
- Industry
- Data Center / Data Center Cooling / Air & Precision Cooling
Our classification, not the publisher's statement. Best suited for, not only for.
MLCommons published the results of the MLPerf Storage v3.0 benchmark suite, which evaluates storage performance for machine learning workloads in an architecture-neutral way. The updated suite adds coverage for additional workload types and an S3 object storage access layer to support comparisons across hosted storage options.
Version 3.0 broadened the tests to reflect a wider range of storage workloads that AI systems can generate, alongside existing support for a POSIX-compliant layer. The round included 19 organizations submitting results, and about one-sixth of total submissions used the S3 storage access layer.
The v3.0 suite expanded test coverage with two new benchmark tests. It added a KV Cache test for LLM inference cache read/write operations. It also added a Vector Database (VDB) test for vector indexing and querying workloads, with support for training, checkpointing, and some VDB tests under the S3 access layer.
MLPerf Storage working group co-chairs Brian Belgodere and Curtis Anderson, and MLPerf head David Kanter, discussed the changes and participation. Belgodere said, “These new additions to the benchmark suite round out the test collection, covering a larger range of AI inference workloads that drive storage needs,” and Anderson said, “Further broadening our support for a diverse set of storage systems, including S3, in the MLPerf Storage benchmark suite gives organizations that are provisioning AI systems even greater ability to select and combine technologies to meet the specific technical requirements of their application.” Kanter added, “We would particularly like to welcome the eleven first-time submitters: Azure, Everpure, HolmesAI, Nebius, NewFW, NVIDIA, OpenLake, Suzhou Zishan Longlin, TuringData, XSKY, and ZettaLane,” and said the storage benchmark represented cloud-based and on-premises solution providers as well as organizations developing storage systems and devices. The release also reported power-efficiency results for checkpointing write and UNet3D read tests, and said the suite was created through collaborative engineering over five years.
Press release, provided by Globe Newswire on behalf of MLCommons. Read the original.