Nvidia GTX 680

[Pizzadude]

Anybody know how this bad boy performs in Seti apps ?

[Mike]

I have checked a few results at seti.
680 is a little faster than 580 but not much.

[Richard Haselgrove]

I've seen the results of some offline testing done with our standard bench testing tools. Tasks validate properly with all the regular applications - specifically, with stock v6.10 and x41g from the installer - so there is nothing to stop the cards being used right away at SETI. We don't have a repeat of the Fermi disaster, thank goodness.

As Mike says, the speed isn't spectacular with the current applications, but there are signs of significant gains with applications specifically written and compiled with the new architecture in mind. As always, it's the combination of hardware and software - especially the software development tools supplied by the manufacturer - which makes the difference, not just hardware on its own.

[MarkJ]

GPUgrid had one donated. They think they will need to release a cuda 4.2 app, so are looking to have a beta app out fairly soon. A bit of a jump for them as their current app has been compiled with cuda 3.1.

I understood Jason had a cuda 4.1 multi beam app he was testing, maybe it just needs a polish and can be released fairly soon?

[arkayn]

I think Jason is waiting for v7 to launch before he releases any more apps.

[William]

GPUgrid had one donated. They think they will need to release a cuda 4.2 app, so are looking to have a beta app out fairly soon. A bit of a jump for them as their current app has been compiled with cuda 3.1.

I understood Jason had a cuda 4.1 multi beam app he was testing, maybe it just needs a polish and can be released fairly soon?

The currently released x41g_cuda32 does work on Kepler, so no rush to get something new out.
Jason intends to put out a more targetted CUDA 4.1 app in due course.

[Richard Haselgrove]

Though he told me that 'in due course' perhaps depends on when cuda 4.2 is released - and he didn't expect that to be available for public distribution until the GPU Technical Conference, May 14-17 in San Jose.

[aaronhaviland]

Though he told me that 'in due course' perhaps depends on when cuda 4.2 is released - and he didn't expect that to be available for public distribution until the GPU Technical Conference, May 14-17 in San Jose.

Odd, I haven't seen anything about 4.2 in the usual places, but an NVIDIA employee has posted links to a 4.2 RC: http://forums.nvidia.com/index.php?showtopic=225731

[glennaxl]

Any preliminary benchmark numbers?

[Pizzadude]

With the current generation of Seti apps the GTX680 is about the same as the GTX580 - however the clever bit is that the GTX680 achieves this whilst consuming 20% less power, good news if you live in the UK where electric prices have shot up in recent years.  Perhaps the next generation of apps will squeeze a little more performance from this baby.

Any preliminary benchmark numbers?