Characterization and Machine Learning Classification of AI and PC Workloads - research and classification methods

Aug 2, 2026

Upload by Asher

Work focuses on profiling artificial intelligence (AI) and standard personal computer (PC) workloads to support AI processor design. AIBench and PassMark PerformanceTest are profiled on a multicore system using the Intel oneAPI VTune Profiler, capturing CPU and platform metrics plus event counts. Orange 3.0 is used to train and test nine machine learning models to classify CPI and elapsed time across benchmark tests, including AIBench training/inference and PassMark CPU, memory, graphics, and disk tests. Linear regression best predicts CPI, while a 4-hidden-layer neural network best predicts elapsed time, enabling workload distinction and highlighting stressed units for future improvements.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages18

Reading Time45 min

AccessFree to download

Abstract

Introduction

Benchmarking background for AI and PC workloads

AI and ML benchmark landscape

Study intent and methodology

Methodology overview

Running AIBench and PassMark

Capturing metrics and event counts with VTune

Training ML models for CPI and elapsed-time classification

Received 24 April 2024, accepted 7 June 2024, date of publication 12 June 2024, date of current version 20 June 2024. Digital Object Identifier 10.1109/ACCESS.2024.3413199 Characterization and Machine Learning Classification of AI and PC Workloads FADI N. SIBAI1, ABU ASADUZZAMAN2,(Senior Member, IEEE), AND ALI EL-MOURSY3,(Senior Member, IEEE) 1Department of Electrical and Computer Engineering, Gulf University for Science and Technology, Mubarak Al-Abdullah 32093, Kuwait 2Electrical and Computer Engineering Department, Wichita State University, Wichita, KS 67260, USA 3Electrical and Computer Engineering Department, University of Sharjah, Sharjah, United Arab Emirates Corresponding author: Fadi N. Sibai ([sibai.f@gust.edu.kw](sibai.f@gust.edu.kw)) ABSTRACT To better design AI processors, it is critical to characterize artificial intelligence (AI) workloadsand contrast them to normal personal computer (PC) workloads. In this work, we profiled the AIBench and PassMark PerformanceTest benchmarks with the Intel oneAPI VTune Profiler on a multi-core computer. We captured and contrasted the various CPU and platform metrics and event counts for these two distinct benchmarks. Using the Orange 3.0 data mining tool, and based on the captured profile metrics and event counts, we then trained and tested 9 machine learning (ML) models to classify the CPIs and elapsed times of the various tests of these two benchmarks, including inference and training tests in AIBench, and CPU, memory, graphics, and disk tests in PassMark. The linear regression machine learning model emerged as the best clocks per instruction (CPI) classifier, while the neural network model with 4 hidden layers was the best elapsed time classifier. This machine learning classification can help in predicting the CPI and elapsed time and distinguish between AI and standard PC workloads based on the profiled application(s) and captured profile metrics and event counts. The stressed computer units identified by this detailed profiling work and exercised by the benchmark tests can also guide future AI processor design improvements. INDEX TERMS AI workloads, Tensorflow, PassMark PerformanceTest, AIBench, workload characterization, event counts, benchmark profiling, machine learning classification, VTune. I. INTRODUCTION The amount of research devoted to artificial intelligence (AI), machine learning (ML), and deep learning (DL) has accelerated. Applications of AI, ML and DL have spanned a wide range of fields including health, engineering, business, agriculture, and arts. A critical factor in successful AI deployment is the performance of the computing infrastructure. Computer benchmarks were created to assess the performance of computers. Computer performance has relied on benchmarks such as 3DMark [1], [2], SPEC [3], PCMark [4], [5], and Open Source Mark (OSMark) benchmarks [6] . These benchmarks attempt to stress the components of computers such as the CPU, memory system, or I/O system. Other performance studies have focused on narrower subjects such as the performance of the cache hierarchy [7], performance The associate editor coordinating the review of this manuscript and approving it for publication was Antonio J. R. Neves . impact of cache locking schemes [8] and thread scheduling and migration [9], or performance comparisons of libraries such as MPI vs Pthread [10], or OpenCL vs CUDA [11] . PerformanceTest 11.0 (PassMark) [12] is a computer benchmark similar to PCMark which measures the performances of the CPU, memory, 2D and 3D graphics, and the hard disk. More recently, the performance of deep learning and machine learning systems has caught the attention of several researchers to characterize workloads, identify performance bottlenecks in hardware and software stacks, and assess the performance gains due to various accelerators. AI benchmarks include DawnBench, ParaDNN, HPL-AI, MLPerf, and AI Benchmark (AIBench) . MLPerf [13],[14] includes seve

FAQs

What benchmarks are used to represent AI and standard PC workloads?

The study profiles AIBench for AI workload behavior and PassMark PerformanceTest for standard PC workload comparison.

How are performance data and event counts collected?

Intel oneAPI VTune Profiler is used to monitor CPU and platform components and capture metrics along with event counts for both benchmarks.

Which machine learning models perform best for CPI and for elapsed time?

Linear regression provides the best CPI classifier, while a neural network with 4 hidden layers provides the best elapsed time classifier.

Aug 2, 2026

Upload by Asher