Predicting Computer Performance Based on Hardware Configuration Using Multiple Neural Networks

Abstract

How can we accurately predict the performance of a Personal Computer (PC) configuration without time consuming simulation? In this work, we predict the performance of a computer hardware configuration using Multiple Neural Networks (MNN). We use Principal Component Analysis (PCA) during data preprocessing as guidance for model creation. The input data includes the internal component characteristics of a computer. A deep learning model is used to infer a benchmark score given a hardware configuration. The finished model takes input data such as Central Processing Unit (CPU) type, frequency, number of cores, memory size and speed, flash or disk architecture, network configuration and correlates it against the corresponding performance benchmark value and system response to a benchmark workload. We demonstrate the accuracy and effectiveness of the MNN and PCA machine learning models using the Standard Performance Evaluation Corporation (SPEC) benchmarks (SPEC CPU2006 and SPEC CPU2017), and a set of approximately 50,000 commercial machines configurations. Our MNN model is able to achieve an average accuracy rate of 97.5% for all benchmarks. Our results provide both personal and enterprise users a tool that can accurately estimate system configuration performance without lengthy and resource intensive benchmarking sessions.