Baidu’s DeepBench to Evaluate Deep Learning Operations

Baidu, Inc.’s (BIDU – Snapshot Report) research unit Baidu Research is working on GPU-based deep learning at its Silicon Valley AI Lab (SVAIL). The unit yesterday announced DeepBench, an open source yardstick to measure the speed of processors in training machine learning models. The benchmark currently tests low level operations like matrix multiplication, handling recurrent layers, convulsions and time taken in sharing data with all processors collectively. DeepBench as highlighted by SVAIL marks the first step toward improving computing performance. With DeepBench, process designers will be able to track processor speed on different hardware platforms and optimize those platforms for deep learning operations. BAIDU INC Price Sharan Narang, software engineer at SVAIL stated that “DeepBench will help organizations — from startups to large companies — identify the optimal hardware platform…


Link to Full Article: Baidu’s DeepBench to Evaluate Deep Learning Operations