G80 Folding?

[XC] gomeler · 03-28-2007, 01:17 PM

I know this would probably be big news if there was even an announcement but.. are there any rumors of a G80 folding client or is Nvidia not working with Stanford like ATi is/was? I really want a client that'll run x86 code on a GPU, then I could run all my projects

justin_c · 03-28-2007, 01:22 PM

they have plans for a nvidia folding client.

the only reason why its been held so far back is because the ATI cards are more adept at folding than the nvidias.

[XC] gomeler · 03-28-2007, 03:39 PM

Damn shame considering how many G80s are floating around the world right now idle. They just need to somehow use virtualization hardware to run x86 code on the GPUs... somehow

Thanks for the info, will keep my eyes peeled for a G70/G80 client.

Sorrow13 · 03-28-2007, 04:21 PM

the issue is (and likely will be) nvidia driver support. the hardware capability is there, but nvidia drivers simply do not work properly with folding.
this has been discussed at length HERE

embeejay · 03-28-2007, 05:38 PM

Quote:

Originally Posted by [XC] gomeler

Damn shame considering how many G80s are floating around the world right now idle. They just need to somehow use virtualization hardware to run x86 code on the GPUs... somehow

Thanks for the info, will keep my eyes peeled for a G70/G80 client.

Gpu's are specialized processors, they aren't able to do a cpu's job, virtulization or not, they simply can't do all the type of work a cpu can. what they can do is a small subset of what a cpu can, but way better than a cpu

couppi · 03-28-2007, 07:38 PM

Last I heard, even though the G80 has 128 unified pixel pipelines or whatever, only 16 at a time can be used for most folding calculations. For some specialized calculations, 32 pipelines can be used. The folding performance would be on par with an x1600 series card at best, and the Pande group doesn't feel that that's a worthwhile contribution

Sorrow13 · 03-28-2007, 09:10 PM

if youre speaking of pixel granularity, i was under the impression that meant only blcks of 16 things at a time can be done. For instance, with 128 shaders, 8 groups of instructions can be issued if they are the right size.
HERE
Is an explanation ofwhat i believe pande group is referring to, albeit a bit old (only refers to 7900 series)
I believe the 8800 series is competetive in that area, if not as completely efficient as the ATI solution, the cards themselves still hold great potential.

couppi · 03-28-2007, 09:28 PM

It took some googling, but I found the thread. On the sixth page, it looks like folding on G80 is more or less a driver issue, but I haven't read the whole thread.

SparkyJJO · 03-28-2007, 10:02 PM

Quote:

Originally Posted by couppi

It took some googling, but I found the thread. On the sixth page, it looks like folding on G80 is more or less a driver issue, but I haven't read the whole thread.

Why does that not surprise me.... nVidia drivers strike again

[XC] gomeler · 03-28-2007, 10:53 PM

Quote:

Originally Posted by embeejay

Gpu's are specialized processors, they aren't able to do a cpu's job, virtulization or not, they simply can't do all the type of work a cpu can. what they can do is a small subset of what a cpu can, but way better than a cpu

Oh trust me, I know

The ALU monsters that are GPU cores just need to be unleashed, rather than calculating vertexes they need to be crunching numbers for me. I would kill for all those gigaflops on a few of my pet projects.

eXceededgoku · 03-28-2007, 11:35 PM

lol teh tables have really turned in the "ati makes sucky drivers" since CCC is actually almost bareable now (:O) and nvidias drivers now suck with a crappy control panel... what has this world come to!?

embeejay · 03-29-2007, 01:41 AM

hehe i couldn't agree more nvidia has taken a turn for the worse.. my next gpu will probably be an amd although i've been sticking with nvidia since the TNT2