ekondis/mixbench

A GPU benchmark tool for evaluating GPUs and CPUs on mixed operational intensity kernels (CUDA, OpenCL, HIP, SYCL, OpenMP)

What it solves

It provides a way to evaluate the performance bounds of compute devices (GPUs and CPUs) by testing how they handle mixed operational intensity. This helps developers determine the practical optimum balance between compute operations and memory accesses to effectively hide memory latency.

How it works

The tool executes customized kernels across a range of operational intensity values. It tests four specific types of operations combined with global memory accesses:

  1. Single precision multiply-additions (Flops)
  2. Double precision multiply-additions (Flops)
  3. Half precision multiply-additions (Flops, GPU only)
  4. Integer multiply-addition operations

Who it’s for

It is designed for researchers and developers working with high-performance computing who need to profile hardware performance and estimate kernel performance bounds.

Highlights

  • Broad Hardware Support: Includes implementations for CUDA, HIP, OpenCL, SYCL, and CPU/OpenMP.
  • Multi-Precision Testing: Supports single, double, and half precision floating-point operations as well as integer operations.
  • Performance Analysis: Generates detailed CSV data including GFLOPS and memory bandwidth (GB/sec) to help build roofline models.

Related

  • Project
    krrishnarraj/clpeak

    A synthetic micro-benchmark for measuring the peak compute performance of CPUs and GPUs across multiple backends like CUDA, ROCm, and Metal.

  • Project
    KernelTuner/kernel_tuner

    An auto-tuning tool for GPU kernels that benchmarks and optimizes parameters across multiple GPU programming languages to improve performance and and energy efficiency.

  • Project
    pytorch/cpuinfo

    A cross-platform C/C++ library for detecting detailed CPU hardware information, such as instruction sets and cache topology, to enable performance optimizations.

  • Dispatch
    Making Deep Learning Go Brrrr: A First-Principles Guide to GPU Performance

    Learn how to identify and resolve performance bottlenecks in deep learning models by understanding the interplay between compute, memory bandwidth, and framework overhead.

  • Project
    NVIDIA/nvbandwidth

    A performance benchmarking tool for NVIDIA GPUs that measures memory bandwidth and latency across CPU-to-GPU and GPU-to-GPU links, including multi-node configurations.