HeteroBench: Multi-kernel Benchmarks for Heterogeneous Systems

Description

HeteroBench is a comprehensive benchmark suite designed to evaluate heterogeneous systems with various accelerators, including CPUs, GPUs (NVIDIA, AMD, Intel), and FPGAs (AMD). It features multi-kernel applications spanning domains like image processing, machine learning, numerical computation, and physical simulation. HeteroBench aims to assist users in assessing performance, optimizing hardware usage, and facilitating decision-making in HPC environments.

Key Features

Broad Hardware Compatibility: Supports CPUs, GPUs from multiple brands, and FPGAs.
Multi-language Support: Benchmarks available in Python, Numba-accelerated Python, and various C++ versions (serial, OpenMP, OpenACC, CUDA, and Vitis HLS).
Customizable Kernel Placement: Enables kernel-level optimization for different hardware backends by users.
Fair Comparisons: Standardized algorithms across platforms for unbiased benchmarking.
User-friendly Design: Simplified setup with configuration files and a top-level Python script for project management.

Requirements

Python3
LLVM/Clang
NVIDIA CUDA Toolkit
NVIDIA HPC SDK
Docker

Website

https://github.com/HewlettPackard/HeteroBench

Contributors

Hongzheng Tian
Rolando P. Hong Enriquez
Dejan Milojicic
Eitan Frachtenberg
Zhiheng Chen
Sitao Huang

Maintainers

Alok Mishra, alok.mishra(at)hpe.com
Hewlett Packard Enterprise

Version

1.0

License

Downloads

Repository

Related publications and projects

Hongzheng Tian, Alok Mishra, Zhiheng Chen, Rolando P Hong Enriquez, Dejan Milojicic, Eitan Frachtenberg, and Sitao Huang, HeteroBench: Multi-kernel Benchmarks for Heterogeneous Systems. In Proceedings of the 16th ACM/SPEC International Conference on Performance Engineering (ICPE 2025).