Mac hardware

Hi, I'm a Mac user and will be buying a new 27 inch iMac for my research lab. I will use Matlab mostly for EEG analysis (EEG/ERP lab). Will there be a significant performance difference between the 3.1Ghz i5 and 3.4Ghz i7 processors for that purpose?

Any advice will be much appreciated.

Thanks!

[Walter Roberson]

Is "EEG/ERP lab" a specific software package? If so then someone would need experience with that package in order to comment. If you are using custom software then we would need to know what kind of operations you are doing and what kinds of array sizes you are using.

Is your work more exploratory ("run this program and then look at the results and figure out what could be changed") or more processing of large numbers of spectra using well-established programs? I ask in order to get an idea of the extent to which multi-core and parallel processing might be important, and also to lead into asking whether you were hoping to do some work on GPU-based processing.

[Daniel Shub]

I think Roberto is referring to the http://erpinfo.org/erplab and http://sccn.ucsd.edu/eeglab/ toolboxes.

[Jason Ross]

Also, don't forget about memory and disk performance if those are important to your application. It looks like the standard iMac comes with 4 GB RAM, which can get gobbled up fairly quickly. SSDs improve disk performance, although they are more costly.

But the fastest processor is going to sit idle if it's got to wait on swap, so these upgrades might be worth it if your application performance could benefit.

[John D'Errico]

A few years back, I would not consider a Mac for MATLAB use that had less than 8GB of RAM, and I have at least a few times pushed into virtual memory even with that. In fact, depending on what you are doing, you may find the money better spent on extra RAM than a faster CPU.

[Walter Roberson]

Unfortunately I do not have any experience with those particular packages. I do see that they are written to handle MATLAB 7.0 onwards, and there is mention of the need for the object oriented facilities, which in 7.0 would have had to have been the "+" directory structure rather than classdef.

I would suspect then that the code was not written with support for the parallel toolbox, and so if it benefits from multi-core support at all, it would be when internally calling the LAPACK and BLAS routines (as those are tuned to use multiple cores if available.) But I don't know if the array sizes would be large enough to for those routines to kick in.

My _speculation_ at this time be that the package is possibly going to be (single-core) CPU bound (and possibly graphics I/O bound as well). If that is true, then I would expect the i7 to have a bit of an advantage... but possibly not enough advantage to be worth the price difference. Too many unknowns to say.

Roberto, is this going to be mostly periods of long processing, or is it going to be "process this chunk of work while the user is sitting there, and then the user will do some relatively light work looking at the outputs before launching another burst of work" ? If it is mostly periods of long processing, then you might be interested in the computers built by Liquid Nitrogen Overclocking.