Has the GTX 680s GPGPU capability been thoroughly explored?

[Vesku]

Where does it rank? There was some talk of "optimizations" for the new Kepler architecture at least in regards to @Folding. Is it possible they might see a large jump after optimizing?

Have yet to see any solid numbers on bitcoin MH/s but that most likely means it is quite underwhelming. Napkin estimators were guessing 110ish MH/s.

Please do not reply if it will be to say "ROFL GPUs are for GAMES."

 

[Olikan]

no, kepler don't have all it's the gpgpu power yet...
there is alot of drivers work to be done

other thing, a 28nm fermi would be better for gpgpu, even better than any BIG-K ...well atlest teoricaly :ninja:

 

[Dark Shroud]

The 680s stink at GPGPU.

http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/17

 

[Edrick]

GTX 580 is still the king of GPGPU (from nVidia).

 

[Vesku]

https://bitcointalk.org/index.php?topic=73681.20

Not looking very promising.

 

[Keysplayr]

AFAIK, Nvidia is working with Stanford on a new F@H client that takes advantage of the Kepler arch. No due date yet.

 

[shadowofthesun]

I don't think it's quite fair to just make the blanket assessment that they suck at GPGPU- their floating point performance is severely neutered, but int performance should be unaffected and they do show promise in some benchmarks. I think it will be a very usage-dependent scenario. One thing is for sure though, they will not excel in compute in general like Fermi did.

 

[Vesku]

Bitcoin is int based and in the forum post I linked it has yet to get more than 110MH/s. A Radeon 5830 does about 280MH/s.

 

[The|Hunter]

^
Yeah you cant make magic with only 8Compute Units.

 

[Jaydip]

Technically Kepler should have Compute capabilities of 3.X.But we need a new cuda version to take advantage of the new functionality.

 

[nenforcer]

Double Precision Floating Point is down from 1/8 in GTX 580 Fermi to only 1/24 in GTX 680 Kepler.

You have to remember that this is really only the midrange card (GK104) to the true GK110 will probably be back up to 1/8 and then we will know the improvements in GPGPU.