MLPerf Storage - submissions opening soon!

[David Kanter]

Hi Everyone!

We want to give you an early heads-up that the MLPerf™ Storage Benchmark, from the MLCommons® Storage Working Group, has been finalized and will be open for submissions on June 19th. 

MLPerf Storage is the first open-source AI/ML benchmark suite that measures the performance of storage for machine learning workloads. The benchmark was created through a collaboration spanning more than a dozen leading industry and academic organizations. It will be an effective tool for purchasing, configuring, and optimizing storage for machine learning applications, as well as for designing next-generation systems and technologies. 

The benchmark does not require the use of expensive accelerators and instead relies on a novel and elegant emulation mechanism that captures the full realistic behavior of neural network training. I’m really happy with the Storage working group for accomplishing this!

Benchmarking serves the industry best when there is broad participation. Please take a look at the MLPerf Storage Benchmark now so you are ready to submit when it opens on June 19th, and please help us spread the word to companies you know who will be interested in participating.

MLPerf Storage Benchmark Timeline:

• June 19, 2023  MLPerf Storage open for submissions

• July 28, 2023  Submissions due

• September 13, 2023 Benchmark competition results are published

Here’s how you can get involved:

• Submit a benchmark result

• Spread the word to companies who will be interested in submitting

• Help us continue the development of the benchmark by joining the working group

Please let me know if you have any questions, or would like to be connected to the working group chairs.

Thank you for your support.

David Kanter

Executive Director

MLCommons