Core i7 upgrade for scientific/matlab computing?

Single core performance between the i5 and i7 in the retina iMac is quite different. The i5 is only 83-84% as fast as the i7.

http://browser.primatelabs.com/geekbench3/compare/1674278?baseline=2739909

So yes, by all means, go for i7 if you have the funds to do so.

PS you should at least get a Fusion Drive setup. If funds allow, go pure SSD. The 3TB drives fail quite easily.

 

Yopu'd probably get better results if you optimized your code Are you going to use an SSD, or a fusion drive?

 

first 128 GB.
https://en.wikipedia.org/wiki/Fusion_Drive

since you stated single core, I don't suppose you have much need for gpu compute?

 

Many MATLAB functions are already threaded, and multi-threading your own code for parallel operation is apparently relatively easy. Take a look at http://www.mathworks.com/discovery/matlab-multicore.html

 

You do realize that an nMP is going to be completely overkill, and the i7 5K iMac is more powerful than the baseline quad core nMP.

 

The difference between the two would be negligible for your purposes.

Having said that I believe for scientific computation single threaded programs are not ideal. It makes more sense to do the computation on the gpu; which is an order of a mangnitude faster than doing it on the cpu.

It is not hard to perform gpgpu compute onto your code; you just need the right libraries and a few lines of code.

 

Speaking of gpu compute, does anybody have figures on double precison FLOPS. I know that the D500 has a marked advantage over the D300 in this regard, but I'm not sure where the m290/m290x/m295x fit

 

I can't give you specific numbers. Typically workstation cards have much better double precision performance than the consumer cards.

 

If your laptop is a Mac also, target disk mode might be an option with the 5k iMac. Not the most convenient option necessarily, but these things boot so quickly now...

 

there is Gputest 0.7's Julia64 and Julia32

my results at fullscreen 1080x1920 (I do have a second monitor...) with a m290x

Julia32: 37509 (fps 624)
Julia64: 3618 (fps 60)

meaning double precision is a tenth as slow.

But in reviewing a D500, the makers of that tool said:

• FirePro D500: 28514 points, 474 FPS
• FirePro D500: 9830 points, 163 FPS

​

http://www.geeks3d.com/20140425/amd-firepro-d500-mac-pro-late-2013-quick-opengl-test/

which suggests that the fp64 speed penalty is not as harsh with the D500. IIRC the D700 is similarly efficient, but the D300 is not very good at double precision math.

(drivers have improved since the D500 review was written)
Of course, all this is irrelevant if you can't parallelize your code.