Hi everyone!

If this thread doesn't belong in this forum, please, move it, but I haven't found a better place to post it to.

Anyway, here's what I was pondering and I was hoping you could help me understand how I could go about implementing the following. I decided to go all the way and make an eight (8) monitor setup capable of crunching data for BOINC projects (such as Einstein or SETI) that also will excel at gaming and take care of computation heavy desktop applications such as CAD or Adobe Creative Suite.

I don't care about the money that much, at this point I'm more interested in finding the optimal solution. Optimal for me would be to have a powerful cruncher, gaming platform and desktop platform in one physical box. Having a peripheral GPU block would be acceptable (referring to Quadro Plex or rack-sized Telsa unit from NVIDIA).

If there's no way to get it all from just one box, as I see it I'd have to make a standalone cruncher but I'm not sure whether I can combine gaming platform and heavy computation desktop applications in one box since those require different adapters. (?)

Any piece of advice would be highly appreciated. This is really boggling my mind because I just don't know how to put all of this together. And it's a rather expensive setup so I'd rather figure things out before I shell out a pretty sum, you know xD

As far as the crunching goes, what would be the best GPU setup? I want to get max results and crunching power :) Tesla cards? Quadro Plex? SLI setup? Multiple cards?

My target OS would be Linux and for what I know ATI's driver support is poor compared to that of NVIDIA, that's why I refer to NVIDIA adapters throughout the post.

Also, any experience with creating 8+ monitors setups? Will this be possible to use with combining CUDA adapters and gaming ones?

Tesla's and Quadros can run in one box, and neither are good for gaming. Geforce cards can't be run with Teslas from what I understand. I see no need for 8 monitors, unless you REALLY need them for something.

Windows Vista/7, and soon 8, and current drivers do not need a monitor attached or a "dummy" dongle attached to use a GPU any more with boinc. Linux is a great OS and good for "some" games. I say some because many games aren't made for Linux. IMO if you want a well rounded crunching/gaming, go with a Win 7 box and use linux for crunching only.

If you want a system that does BOTH crunching and gaming, I'd get a nice mobo with 8PCIe 2.0 slots... and shove them full of dual GPU cards like the nv590.. you won't get the power of teslas, but 8 590s will do quite a bit of work :)

Figure a TOP end nv gpu like a 590 costs about $600 and, it's a x2, so you get about 2Glops or so from it (SP math, about hald that for DP).... take a low end Tesla for $2000 and get 1Tflop to 700Glops from it.. and well.. you can see a Telsa card is a MUCH better value if you are JUST running science, etc apps... plus Teslas work with Quadros as well.

now 8 teslas (and 16+K USD).. will scream and give you ~8TFlops of GPU power. no gaming video card can even come close to what a tesla can do.

the BEST gaming gpu, right now, does, I think about 1-1.5Glfops tops. and a low end Tesla does 1TFlops single precision and 700GFlops double

I have a Telsa card, and it can do things in a FRACTION of a time a 580, 590, etc can push out WUs. Just 1 of those beasts is better than a room full of nvidia and ati GPUs.

"the BEST gaming gpu, right now, does, I think about 1-1.5Glfops tops. and a low end Tesla does 1TFlops single precision and 700GFlops double"

For single precision, you are underestimating the capabilities of modern consumer-grade GPU's by a factor of several thousand.

I have a pair of GTX 460's that, according to BOINC, are running the milkyway@home app at 59.67 GFLOPs/sec apiece. They're at least a year and a half old, if I recall, and they weren't particularly high-end when I got them.

Agreed that the Telsa's are mighty badass -- but not 3 orders of magnitude more badass.

Jim

Ok I know old thread, but seriously "David".

1. you cannot run more that 4 cores in SLI, NVIDIA does not support it, you can add a 5th core either by a 5th or 3rd card depending on if you using dual core cards or not, for your physx processor, but it should be very close to the specs of the parent cards or better, or it may actually hurt you physx processing capability.

Now simply for folding you could possibly run a full 8 cards, just not all in sli, which would detract form it gaming ability, now you could mix and run some in sli and the rest not. You could also run 2 dual core in sli and 6 Teslas for crunching.

Teslas do not have display adaptors, so you would still need a normal gfx card for a display, and running 2 680's I believe you can max out at 3 displays because of the cards disabling some ports when using surround vision. could be wrong if their is a tweak to get around that.

2. a Geforce 680 actually performs at 3090 Gflops, yes three thousand and ninty, the 580 does 1581 Gflops, the NVIDIA Tesla C2075 handles 1.03 Tflops.
Now those are spec numbers, crunching results may vary based on OS, environmentals, and other contributing factors.

So please don't attempt to give out advice if you have little working knowledge of the information your attempting to spew as passable.

Put it this way, I registered just to let you know how much your disinformation offended me and wanted anyone else that happens across this to know to pay it no heed.