MLPerf Storage v1.0 Benchmark Measures Performance of Storage Systems

By Amber Ankerholz

Increasing demands challenge existing storage systems.

MLCommons has announced results for its MLPerf Storage v1.0 benchmark suite, which measures storage system performance for machine learning (ML) workloads in an architecture-neutral, reproducible manner. In other words, the company says, the benchmark measures how fast storage systems can supply data when a model is being trained.

The MLPerf Storage benchmark includes three models – 3D-UNet, Resnet50, and CosmoFlow – and emulates storage demands across various system configurations, including different accelerators and workloads.

According to the announcement, the results show that “as accelerator technology has advanced and datasets continue to increase in size, ML system providers must ensure that their storage solutions keep up with the compute needs.”

As MLCommons states, “storage system architects now have few design tradeoffs available to them: the systems must be high-throughput and low-latency, to keep a large-scale AI training system running at peak load.”