AMD GPU benchmark using OpenCL
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
.gitignore
LICENSE
Makefile
README.md
benchlib.c
benchlib.h
cl-info.c
display.plot
main-cpu.c
main.c

README.md

AMD GPU OpenCL Benchmarking

The code in this repository allows benchmarking an AMD GPU's computing power using the OpenCL framework against a CPU.

Installing prerequisites

The followings need to be installed in order to have an OpenCL setup working with an AMD GPU:

Running benchmarks

Provided that the OpenCL library has been properly installed (and the Makefile possibly adapted), make builds the following binaries:

  • cl-info provides information about the OpenCL hardware on the machine.

  • main performs the actual GPU benchmark using the OpenCL framework. It generates the bench_gpu_loops.txt and bench_gpu_workers.txt files. They contain the time elapsed in nanoseconds (right column) for computing a given number of operations (left column), respectively for a single worker and for many workers.

  • main-cpu performs the same computation on the computer's CPU than the GPU benchmark tool. This allows comparing the CPU's computational power against the GPU's. It generates both bench_cpu_loops.txt and bench_cpu_workers.txt files, which kind of translate the benchmark performed on the GPU, though on the CPU. There is almost no difference between these 2 files, since the computation is performed serialy anyway on the CPU.

  • gnuplot display.plot displays a graph and prints some statistics from results generated by main and main-cpu.