0
À l'étude

Performance comparison

Michal Kvasnicka il y a 4 ans mis à jour il y a 3 ans 8

I propose to perform update of the whole MCT performance analysis (paragraph Performance here) with latest versions of MATLAB (2021a_update2), Mathematica (12.3) and Maple (2021.0). I am wiling to perform this analysis, because I have an access to latest version of mentioned SW packages.

I spent some time to discuss with MathWorks guys current "terrible" performance of VPA (Matlab 2021a). So, I thing that is a suitable time to update this comparison and show what is the current performance of MCT together with comparisons.



Solution

Solution

Sorry to hear about miscommunication with "big boys". But I am not surprised at all, as my experience with them was the same. I communicated with several top technical people, they tried to extract deep technical information about toolbox internals (and why it is so much faster), but never provided anything in return.  Later on, I saw they tried to implement some  of toolbox's technical "know-hows", but without much success, as I didn't reveal all the details :).

In any case, thank you very much for your time and efforts. Hopefully we can return to this topic later on.

Pavel, let me know if it is a good idea from your point of view.

À l'étude

Why not? I think it is great idea. Also it has sense to extend the tests to cover speed-comparison of FFT and EIG, SVD for special  types of matrices (e.g. banded, Hermitian, complex).

 

In some if these cases quadruple precision computations in toolbox are faster (at least it was in 2016) than double in MATLAB (in particular solving the banded matrices: https://www.advanpix.com/2016/10/20/architecture-of-eigenproblem-solver/).

The only show stopper for me is (and always has been) significant time and efforts needed to conduct such benchmarks.

I would be happy if somebody would do this (and would help as much as I can). The only suggestion is to use some good many-core CPU, as parallelism is crucially important for matrix computations and FFT.

And, btw, on different platforms performance of all the software packages are different. 

Of course, I would prefer comparison on Windows, as this would be informative for ~80% of users of toolbox.

OK ... so do you recommend to use your currently available scripts? Can you send me the whole package of all available scripts so that we don't miss anything important?

I am using 8-Core Intel Core i9-9900 with 64GB RAM (Windows 10 Pro).

I gave up this activity, because I asked for help the Maplesoft and Wolfram support few times to solve some benchmark Maple and Mathematica scripts problems (optimization of the benchmark code to avoid bad programming habits) and the response was very vague or directly unfriendly.

So, sorry for premature promises...:)

Hi Michael, 

I've done FFT benchmarks for MATLAB R2017b + Advanpix (version 4.4.5.12699) vs. Mathematica 11.2 vs. C + MPFT library back in 2017-2018 using 

4 core Intel Core i7 3770K @ 4.5Ghz, 10-core Intel Xeon E5-2690v2 @ 3Ghz and 20 cores among two Intel Xeon E5-2690v4 @ 2.6Ghz 

If interested, you can take a look at my results at https://www.researchgate.net/project/Multiprecision-FFT-benchmark

Sorry for the blog format, but It's what I have. 

Thank you for very interesting FFT benchmarks. Especially the MTC parallel scaling is very impressive.

MTC is permanently the best available high-level language solution for multi precision computing. 

Solution

Sorry to hear about miscommunication with "big boys". But I am not surprised at all, as my experience with them was the same. I communicated with several top technical people, they tried to extract deep technical information about toolbox internals (and why it is so much faster), but never provided anything in return.  Later on, I saw they tried to implement some  of toolbox's technical "know-hows", but without much success, as I didn't reveal all the details :).

In any case, thank you very much for your time and efforts. Hopefully we can return to this topic later on.