Mac Pro vs. rMBP for Matlab/Simulink simulations

[hajime]

Hello, I read that Apple updated the Mac Pro few weeks ago. Exactly what were being upgraded? I am considering whether to get the rMBP, Mac Pro or both. (I am planning to retire my MBP 2010 17" due to the heavy weight.)

In terms of overall performance and also specific performance in running Matlab/Simulink simulations, how are the following machines ranked?

1x 3.3 GHz 6-core Mac Pro
2x 3.06 GHz 6-core Mac Pro
2.6 or 2.6GHz Quad-core i7 rMBP

Few years ago, it was better to get a faster, less-core Mac Pro than a slower, more-core one. Is this rule still the same?

Thanks.

 

[deconstruct60]

Few years ago, it was better to get a faster, less-core Mac Pro than a slower, more-core one. Is this rule still the same?.

With or without the Parallel Computing Toolbox ?

http://www.mathworks.com/products/parallel-computing/builtin-parallel-support.html?nocookie=true

Likewise just leveraging multi-threaded features ?

http://www.mathworks.com/discovery/multicore-matlab.html


If just poking around with the Student version, then it probably pays to do "GHz" drag racing because limited in feature set. That is not necessarily a good approach for the full product with the additional purchased components.


In your comparison of models the significant Mac Pro update is the 3.06GHz model. The price is different on the Hex core model but if throwing out buying the 3.06GHz model seems likely you are not price sensitive. If you are price sensitive then that's probably part of the parallelism issue. (both in hardware and in software licensing. )

 

[Boomhowler]

MatLab is much better at utilizing several cores now than it was a couple of years ago, that said it still depends a lot on which algorithms you run if more cores or more GHz will be better. Or even more RAM.

If you want to use the GPU-parts of parallel computing toolbox (it works on osx as well nowadays, if I remember correctly) then the non-laptops are much preferred.

If you look at GeekBench scores (which is an ok benchmark when comparing to MatLab tasks) you get the following results:

3.33GHz 6 core: 15574
3.03GHz 12 core: 25375
2.7GHz rMBP: 13260

This should be taken with a grain of salt, but by comparison it should probably be correct. Remember that GeekBench does not take into account disk speed or GPU power. For GPU results, check out Unigine Heaven etc.

So, what are you looking for? Bang/$ or simply just power? You decide

 

[waynefisher]

The GPU parts of the Parallel Computing Toolbox suck. They're a poor imitation of what AccelerEyes has with Jacket for making MATLAB code faster.

I've been using Jacket on my various Mac machines for a couple years now, I do a lot of heavy monte-carlo simulations. Here is a good summary of the comparison between the too: http://accelereyes.com/products/compare