Torben's Corner: Jacket Benchmark Tables
From Jacket Wiki
This page is a new way of representing the performance of Jacket in a collective table. Earlier, Torben's Corner used a Jacket Performance Index based on the study of very detailed and involved experimental work on a large number of computer platforms. Unfortunately, it has (some might say that it was obvious) become too big a task for me to handle in a satisfactory way. I have therefore changed the way the benchmarking of different functions is done.
I have designed a number of benchmark functions, which tests a large number of Jacket/MATLAB functions for vector/matrix and for single/double precision. This benchmark suite further generates a table in Latex format such that it can be used directly in a report or whatever you may need. The source code is obviously open source for everyone to use as long as proper reference is made.
This way of testing does obviously not reveal all the details from the figures I presented earlier. But they should give a good idea of the performance to be gained for a given platform anyway. The Jacket Performance Table is made as a Toolbox or library, which is easy to extend. The source code is collected in a Jacket_benchmark.lib.zip. Just unpack the zip file, and you are almost ready to go. Copy for example the run_Tesla_C2050.m file to run_my_GPU.m and change the necessary strings etc. in the file. After that you just need to execute run_my_GPU and the benchmark will be conducted. There is detailed help if you run the file RUNME_Jacket_Benchmark.
Benchmarking Results
The benchmark has been run on a number of different platforms and you can see the results here:
- GeForce GTX465 GPU and Core i7-975 CPU.
- Tesla C1060 GPU and Core i7-975 CPU.
- Tesla C2050 GPU and Core i7-975 CPU.
- Quadro FX3800 GPU and Core i7-975 CPU.
Download Jacket_Benchmark_100.lib Library
The benchmark library is provided as source MATLAB/Jacket files for all to use. Please use proper reference is you use it. Download HERE.
Go Home: Torben's Corner
