Computer for science and 3D modelling

[guff]

Hey guys.
I have seen so many well put together computers being suggested on this site, and for several different purposes. Being a very un-savvy when it comes to the world of computers, I would like to ask the community for help.

We are buying new equipment in the lab, and with it comes some new computers. Now, we are quite restricted in the brand (due to bureaucracy etc), but we can ask for upgrades to certain components given it stays within our price range (~2k euros).

The tasks that will be ran on these computers are multiple simultaneous analysis softwares (matlab, BESA, etc) with large datasets open, as well as simple, interactive 3D models of brains for visualization and source reconstruction.

We have come to the conclusion that memory (8gb+) and processing power (i5 or i7?) are the most important things for us, as well as a fast and roomy hard drive, whole getting a simple graphics card.

Do you people have any input in this case? Are we focused on the right components? And are they beefy enough? What are some minimum specs, do you reckon?

Any feedback is much appreciated!
Guff

[fixx]

You should first check out if the programs you use support either CUDA or OpenCL GPU rendering because if that's an option you'll get significant benefit out of beefy graphics card.

i7 is definitely the way to go, 16GB of RAM is not overdoing it and depending on size of your material maybe ever 32GB. HDD speed doesn't matter much, but having discs RAID1 mirrored for extra safety might be worth thinking in some cases.

[Remilia]

Xeon with ECC RAM would be a consideration actually. Memory amount is highly dependent on what your programs uses as stated by fixx.

[Akainakali]

As Fixx said, good GPU card can dramatically increase your performance for these sorts of modeling tasks, far more than the CPU. A GPU is a MASSIVELY parallel processing chip due to the demands of 3D modeling and rendering.

However what card you need is very heavily driven by what the software you use is. Since not all of them support GPU acceleration on any card and AMD vs nVidia can be an order of magnitude difference in the performance benefit depending on what is supported.

AMD in general offers better GPU acceleration on consumer level cards, than the consumer level nVidia cards do. The nVidia quadros can be exceptional cards for this, but often have price tags to match and they have typically crippled the GPU acceleration for anything requiring double precision math on the consumer cards. I'm not certain what kind of math you are doing for those applications.

Matlab appears to require CUDA for GPU acceleration, which is specific to nVidia cards. Not familiar with BESA though and nothing was turning up with regards to a quick google.