Memory Bandwidth

Command-line benchmark for comparing memory copy implementations.

Memory Bandwidth measures several memcpy() implementations and exports plain-text tables, with optional Gnuplot charts when Gnuplot is available.

  • C++ benchmark
  • Windows x86/x64
  • AVX/AVX2
  • Gnuplot output
  • zlib license

Overview

Memory Bandwidth Benchmark compares different memcpy() implementations, including the standard compiler implementation and optimized assembly versions.

Results are written as formatted ASCII tables. If Gnuplot is available, the tool also generates PNG charts from the benchmark data.

Screenshots

Memory Bandwidth Intel i7-4960HQ
Memory Bandwidth Intel i7-6700HQ
Memory Bandwidth AMD Ryzen 2400G
Memory Bandwidth AMD Ryzen 5900X

Details

Benchmark Modes

  • Command-line tool configurable through parameters.
  • Uses a 256 MB memory buffer.
  • Runs fixed-read, buffer-read, fixed-write and buffer-write modes.
  • Displays CPU name, cache information, selected mode and memory buffer details.

Copy Implementations

  • Standard compiler-provided memcpy().
  • MMX, SSE and AVX implementations in the 32-bit binary.
  • Intel optimized AVX and AVX2 versions.
  • Agner Fog asmlib.

Platform and Build

  • Windows 32-bit and 64-bit binaries are available.
  • Compiles on Linux with the included makefile.
  • Can compile on macOS x86, although a makefile is not provided.
  • Developed in C++ with Visual Studio 2019.

Download

Latest version: v3.10 (29-March-2021)
ChangeLog | Roadmap | zlib license

Binaries
32-bit and 64-bit binaries for supported operating systems.
Source code
Source code and project files for VS2019+ and GNU C/C++ 4.x.
Previous releases
Get access to all releases repository.