• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Advice for scientific (i.e. non-gaming) build

homesalad

Junior Member
Howdy, I recently came into some National Science Foundation grant money for some work in computational evolution. I hope to spend about $3000 or so on a machine that won't, unfortunately, be used for too much gaming, but will be used for lots of simulation and numerical analysis work. The simulations often run for 1-2 days, and basically involve millions or billions of floating point operations (but not with particularly large data sets, so really hast hard drives aren't a big issue). Stability and cpu speed are the biggest concerns. I'll likely OC the cpu as much as I can without spending too much time tweaking memory timings, or doing anything that might result in long-term stability concerns since the cpu is likely to be close to 100% utilized for days at a time. I often run many simulations at a time, so the more cores the merrier.

Here's the question: Would it be best to go with the familiar
core i7-920
Tuniq tower 120 (or something similar)
ASUS P6T deluxe
6-12gigs ram
WD Raptor/velociraptor
Corsair 620HX power supply

...or would it make more sense to venture into the server world, and try to find a dual socket mobo on which to run a couple of nehalem xeons ? These look expensive, and I don't know much about them, but maybe someone knows of an obvious choice here?

.. or is there a third option for pure number-crunching power?

I've got cash to burn, but I don't want to throw it away, so if there's something that's more expensive, but worth it, I'd be interested.
Thanks for your help!


 
If you're running a lot of concurrent simulations rather than one big multi-core program, why not consider more than one slightly cheaper machine? Rather than wandering into the world of Xeons, think about how cheaply you could put together a low-end Core2Quad box, compared with the high price of motherboard/DDR3/etc for a Nehalem/multi-xeons. Would just be a matter of remote-desktop-ing into each box from your development machine.

In terms of raw CPU horsepower I'd wager you'd get more for your money that way. I wouldn't like to comment on energy bills though 😉


 
the i7 965 system i put together recently cut c compiling time in half from the e8400 systems we have, would highly recommend going i7. If you want some crazy processing power check out the nvidia tesla add in card, it looks like Tiger Direct is about the only place that sells the bare card.
 
Thanks for the ideas - I like the idea of putting together a couple of relatively cheap machines. I think I'm going to stick with the i7-920 since they're not too expensive, have 8 virtual cores, and are pretty quick, but maybe I'll scale back the ram and cpu and put 'em in a couple of cheap cases, or something. A Tesla (or maybe just a couple of GTX 285's) is something I'm looking into, but I'd have to rewrite a lot of code to utilize a GPU efficiently, so it's not too high on the priority list.
 
Put some of the money into a good backup strategy, like 2 x external HDs (one to keep off-site) and possibly a blu-ray burner if you have too much data for DVDs.
 
For pure scientific number crunching I'd say get yourself some PS3's and network them together. Those Folding@Home benches are pretty stupid. Of course I have no first-hand experience on how to go about this in the real world. Maybe someone else does. (Google it?)

I would think the Cell CPU in the PS3's would still trump the Core i7. Now if you could just utilize the RSX GPU that would be icing on the cake. If that is a "viable" solution for you then I think that would win hands down.

If you're expertise is within the mortal realm (like me), then the practical solution, in my opinion, is a straight i7 build as you have listed. I think the server route with dual CPUs might be bandwidth starved with two i7s but honestly I haven't done my homework there.
 
Back
Top