MLPerf's objective is to build a common set of benchmarks that enables the ML field to measure system performance for both training and inference from mobile devices to Cloud services. The organisation is a consortium of researchers from prestigious universities in the United States along with leaders in ML such as Baidu, Intel, Google and others. The recently released MLPerf v0.5 benchmark is the industrys first objective performance test for training algorithms to perform tasks such as computer vision, language translation, personalized recommendations, and reinforcement learning.
"We have demonstrated that WekaIO Matrix delivers superior performance benefits and is uniquely positioned to be the best storage solution for ML workloads", stated Barbara Murphy, Vice President of Marketing at WekaIO. "Our customers are using Matrix to accelerate ML workloads at scale. Matrix is the only file system today that can keep GPUs and CPUs saturated with data, eliminating the IO constraints commonly associated with legacy storage systems. For successful ML outcomes, it's critical for organizations to consider the underlying storage infrastructure. So, we are honoured to contribute our knowledge and proficiency to the MLPerf benchmark which, when established, will help organizations make better informed infrastructure decisions."
In addition, WekaIO participates in the well-established SPEC SFS benchmark - a standard for high-performance computing (HPC) performance - and a proponent of the importance of delivering performance standards for the ML and HPC industry at large.