Nvidea's Tesla C870

[Steve Lux]

Available for just over $1600 from Digitan Technology, but I have to wonder if it is compatable with our crunching projects.

http://www.google.com/products?hl=en...&um=1&ie=UTF-8

[NeoGen]

Those are very high end dedicated graphics card for very specific purposes... I'm thinking that so far only 3D rendering apps and a few others can make use of that kind of power and onboard memory.

But if the Folding@Home or other projects teams would compile clients to run on it, I'm sure there would be crazy people capable of buying one for crunching.

[Steve Lux]

The crunching capacity of 10 computers for the price of a mid level computer. I'm more than crazy enough if it works.

[NeoGen]

Don't know if it would be that much of a ratio of performance, last I heard about the Folding@Home GPU clients, they used alot of cpu power while working. (Never seen it myself though, and haven't been on top of the subject in a while)

So... if it still checks out, then a monster graphics card like that would need something like a quad-core with all cores at full power on it to take the maximum performance on Folding@Home?

[Strongbow]

Just to 'set the record straight' the Tesla (and ATIs Firestream) range are not graphics cards (they do not have display outputs) but they do utilise GPU technology for computation. Think of them more like a mass of coprocessors or physics engines.

The problem with the Nvidia ones at the moment for my business (finance) is that they are still single precision, but apparently the G92 based ones will change that next year. The ATI Firestream can handle double precision but they drop well below the 500GFlops rating they quote for single precision.

The challenge is whether they will be as easy to develop for as the ClearSpeed cards, I'm sure they will be in time but they are very much in their infancy at the moment so it is worth just monitoring how they mature over the next 12months.

Now the big question is; will someone code a crunching app for them? Don't see why not as it shouldn't be too complex. It would rather be down to whether or not they feel the general public would purchase such a card just for their crunching effort as they are not general purpose cards which could be used for typical public use (games etc.).

[Brucifer]

Whether or not some crunching phreak purchases one, along with the SDK will boil down to bang for the buck. And how long will it be around before the technology changes, etc. At the price, a real penny pinching person could get close to 3 quads going. What's the performance comparisson going to be, with the quad already having clients out there. And as intel brings out more stuff, the cost of quads will drop too. It's not in the same category as the ps3 I don't think where you have a large amount of processing power with a very reasonable current draw, and the ps3 being in a relatively cheap price range too. So it will be interesting to see, but myself, I'd spend the bucks on quads at this point in the game. Maybe in another year or with the next generation of the cards, thing will change, but it's a bit early in the game at the moment I think. Not that that's worth anything.

edit: also going the quads route, one doesn't have all their eggs in one basket either.

[Strongbow]

You actually have to buy around 15 Intel Quads to match one of these cards and when one of these cards only uses roughly 170w you start to see the benefits they could have for businesses that need to handle complex algorithms.

I have to agree with Brucifer though; you're much better off buying PS3s or Quads if you want to crunch.

[Brucifer]

Well I'll be the one to step on it here... But, if one of these cards can put out so much work, at a mere 170watts, then what's the biggie here over the standard cpu's?? ie, why waste more time money and effort on the cpu world when the card is more efficient?

Or it is that it is limited in what you can crunch with it like the cell?

[Strongbow]

You still need a general purpose processor to control it but these cards are far more efficient at handling parallel (or massively multi-threaded architecture) algorithms. The CPU would handle the fetch and forward and the GPU handles the execute which pretty much negates the use of the CPUs ALU (which is really only for basic computation anyway!).

In time we may see that the traditional x86 CPU evolve into a chip which has some of its components decoupled from the silicon for general purpose numerical computation but for now at least we have to write specific proprietary code that handles a particular vendor's card.

It's a watch this space type technology (whilst crunching away on our quads and PS3s )