Nvidia, DataDirect Networks Shine in Latest Inference and Storage Benchmark Results

Since ChatGPT’s launch last November, much of the discussion around artificial intelligence has shifted from awe and amazement to more pragmatic questions such as how to manage exploding compute costs associated with developing and using it commercially. 

Luckily, new findings from MLCommons, a non-profit that develops evaluations and datasets for machine learning, may help answer such questions. On Monday, the organization released a benchmark comparing chips and software that power AI models to make predictions—which is what happens when ChatGPT spits out an answer, known as inference. A second MLCommons benchmark measured how quickly data was moved from different storage systems to the server chips that use the data to train AI models. It’s an underappreciated component of AI model development.