Benchmarking the performance and energy consumption of the AVX512 and VNNI instruction sets

Abstract

This project is focused on measuring the execution time, the energy consumption and the performance of the new instruction set introduced by Intel in the Cascade Lake series of processors, which are called Vector Neural Network Instructions (VNNI). These instructions are part of the AVX512 instruction set, and they are specifically designed to accelerate deep learning codes. To analyse the performance of these instructions, a set of benchmarks will have to be designed and developed. In addition, the impact of using these instructions inside HPC containers will be also evaluated because HPC clusters are the natural place to use these high-end architectures