Benchmarking Data Analysis and Machine Learning Applications on the Intel KNL Many-Core Processor

摘要

Knights Landing (KNL) is the code name for the second-generation Intel XeonPhi product family. KNL has generated significant interest in the data analysisand machine learning communities because its new many-core architecture targetsboth of these workloads. The KNL many-core vector processor design enables itto exploit much higher levels of parallelism. At the Lincoln LaboratorySupercomputing Center (LLSC), the majority of users are running data analysisapplications such as MATLAB and Octave. More recently, machine learningapplications, such as the UC Berkeley Caffe deep learning framework, havebecome increasingly important to LLSC users. Thus, the performance of theseapplications on KNL systems is of high interest to LLSC users and the broaderdata analysis and machine learning communities. Our data analysis benchmarks ofthese application on the Intel KNL processor indicate that single-coredouble-precision generalized matrix multiply (DGEMM) performance on KNL systemshas improved by ~3.5x compared to prior Intel Xeon technologies. Our dataanalysis applications also achieved ~60% of the theoretical peak performance.Also a performance comparison of a machine learning application, Caffe, betweenthe two different Intel CPUs, Xeon E5 v3 and Xeon Phi 7210, demonstrated a 2.7ximprovement on a KNL node.