Performance of Caffe on QCT Deep Learning Reference Architecture — A Preliminary Case Study

摘要

Deep learning is a sub-set of machine learning practice employing models based on various learning network architectures and algorithms in the field of artificial intelligence. Businesses planning to adopt a deep learning solution should comprehend a set of complex choices in hardware, software, configuration and optimizations to setup a functional deep learning solution. This paper will describe the reference architecture built on Intel Knights Landing processor and omni-path interconnection. We provide a simplified guide to deploy, configure and optimize deep learning solutions based on an array of compute, storage, networking and software components offered by Quanta Cloud Technology. The performance data is presented and it shows good scaling and accuracy on processing the data from IMAGENET.