Skip to content

3 The Gaussian Benchmark

Jump to bottom
Thierry Dumont edited this page Mar 7, 2019 · 24 revisions

Here we compare genuine implementations, but also what happens when we call library routines.

This page is under construction.

  • Results on the reference machine, not using libraries:

Without libs

  • Results on the reference machine, using libraries:

With libs

Comments:

1- OpenBlas whas used multithreaded. So the results for C++Lib/ are largelly improved by the parallel computation on 4 cores on the reference machine. The same is true for JuLib/ and PyScipy/. Note that the improvement for PyScipy/ and C++Lib/ is larger than 4 for large sizes, which just shows the efficiency of OpenBlas lib (when compared with a naïve programming). This is not true for JuLib/. Why?

--- to be continued ---

Clone this wiki locally