Researchers Introduce Deep500 Benchmark

New paper provides detailed analysis of deep-learning benchmark for HPC systems

Researchers at ETH have unveiled a detailed analysis of the Deep500 deep learning benchmark, which was introduced at SC18 last November. Deep500 is an attempt to provide “the first distributed and reproducible benchmarking system for deep learning.” The goal of the project is to offer a fair means of comparing “deep learning frameworks, algorithms, libraries, and techniques.”

Deep500 developers believe that conventional HPC benchmark systems are not well suited for measuring and comparing deep learning workloads. As the name indicates, the Deep500 is intended to serve as a universal means of comparing HPC systems, much like the Top500 and Green500, and there is already talk of starting a Deep500 list of the worlds fastest deep learning systems.

See the article at HPCWire for additional information on Deep500.