Indian_Engineer

since I do research in computational biology and love to make algorithms I want to build my own supercomputer using CUDA, custom Linux OS i will make using LFS, and 2 9800GX2's (not in SLI)

this will essentially be a little less than half the computational power as the one posted in news that a uni in Belgium made

I GOT (good deals )
5000+BE $70
BIOSTAR TForce TF570SLI $50
8GB OCZ ram $160 (bought 2gb sticks for $40)

WILL GET
2 9800GX2's. which brand? i was thinking PNY, or eVGA
rest of stuff in sig/plan
4 raptors in RAID10 sounds good. any good deals on raptors, or just newegg?


MAJOR EDIT: CPU power is INSIGNIFICANT WITH A BUILD LIKE THIS SO DON'T FLAME FOR NOT GOING INTEL I only plan to oc to 3GHz just 'cause

Manyak

Why don't you use the same motherboard they did, with 4 x16 slots? This way you have room for expansion if you need it.

__________________
Under Construction: [Project] Wraith
On Hold: [Project] Belial

nitteo

Why dont you email nVidia, they are banking that CUDA will take off. You never know, they might sponsor you with cards.

__________________
Life is either a daring adventure or nothing at all.

Help cure Cancer and FOLD for TEAM OCN.

nickbunyun

LFS? live for speed?

Compaddict

I always buy my video cards from XFX because of the double lifetime warranty. I'm glad too because Nvidia folding just may wear them out eventually.

dankoni

Changing the Raptors to 4xWD6400AAKS would save you a lot of money. Maybe enough for another GX2 =)

__________________

cognoscenti

and I avoid XFX like they plague since they think its acceptable to say 'we no longer do Ultras here have a GTX'....

__________________

Q6600 @ 4017.37 MHz on Striker II

Q6600 3802.47 MHz (422.5 * 9) 790i Ultra

Q6600 4152.04 MHz (461.34 * 9) 8-8-8-18 790i Ultra

trueg50

Linux From Scratch.

Sounds great!

However I would recommend a board with support for at least 3 PCI-E slots. Maybe 680i or something.

SLI doesn't matter in the least bit, however you will need more PCI-E slots to expand.

Also, it might work better to have a single 280 or 260 GTX, instead of a couple 9800GTX x2's.

__________________
Tygon Tornado build pics; at post 23 / How to Cable manage your Thermal take Armor/E6400 3.6 ghz

cuy50

Keep price in mind.

Edit: CD is selling a couple raptors
http://www.overclock.net/cyberdruidp...rogress-3.html

Edit 2: Also you might want to consider two VelociRaptors unless you are dead set on Raid 10.

__________________

FearMeansControl

As to your major edit, I wouldn't say that CPU is insignificant in CUDA..

Only problem with multiple GPU's (and the GX2 as well) is that in the current incarnation of CUDA, data still gets pushed back through the host in order to get from GPU to GPU.. This doesn't mean anything if you aren't writing multi-threaded code, but from the talk of going for two GX2's, it sounds like you probably will. The more data you send back and forth, the more CPU will be needed to push 'stuff' through the pipeline.

I personally think AMD is a better choice here for the memory controller, but that advantage seems to be diminishing year by year, and by December, Intel may be the path to go.

As a side not, throw an Antec Spot Cool over that RAM when you build.. 4 sticks being used for something other than Firefox produces a nice amount of heat.. that your A64 won't like

__________________
Q6600
(450x8) (417x9) (425x9)