GPU Folding - R600 & G80

[Kougar]

While there is not much going on here, I've gleaned a few things might interest a few of our folders here and anyone else that stumbles across this thread. After a careful perusal of the Folding@Home forums I can draw two conclusions regarding GPU client folding regarding G80 and R600.

1) Bad news first: Folding on the G80 appears to not be going anywhere. There are some issues due to how nVidia implements DX Support on these cards, so easily porting over the ATI client won't be possible. Currently support appears to be more a question of upcoming driver changes more than anything else, but I am not entirely sure. So for everyone wondering when they could folding on their 8800GTX or 8800GTS, I'm afraid the answer is not anytime soon. :cry:

2) Good news: It appears that there will be easy porting of the ATI GPU client to R600 when it launches, and as a consequence ATI users will be folding shortly after R600 finally launches, whenever that might be. Infact reading between the lines, one of the GPU clent team appears to already have received early hardware by ATI for exactly this reason... :smile_evil:

[Gray]

Hmmm to SETI or not to SETI.....

Next year I will be deciding again. With the new system, by 1 year around this time It may very well be a diffrent ballgame. I may very well may go back to SETI.

I can't wait to see what I get.
I have always been an ATI fan but I have also been an Intel fan, both exclusivly for my main gaming rig.
We will see what happens next year... I am keeping a close eye on the programable GPU game in the mean time.

[Kougar]

From what I read SETI is in the same boat as the F@H people. The difference between F@H and SETI@HOME is that SETI announced a project for G80, but it's "still in the works" and assumed to be stonewalled. Notice that this is partly speculation, they might release a SETI G80 client tomorrow. ;)

[Raptorfury]

i am a folder myself . and i am wondering how much you gain by using the gpu client over the normal client ? oh yea i am using a ati x1950pro .

[Kougar]

A full 48 shader equiped Radeon will outperform a 3.8ghz Core 2 Duo in points, even if the Core 2 Duo is folding mostly bonus WUs.

That said, the actual WORK difference is much greater. The graphics card actually completes a great deal more work. The figure is out there, I believe the rough # was 4x the work. So if you are folding for the cause and not just points, then GPU folding is where you would want to be.

[liqnit]

What i dont get is why is so hard to work on the G80.
Nvidia claims this GPU is the first to be GPGPU(General Purpose) and also release a language called CUDA - whiche allows developers write a C like code into the GPU.

[SonicWRX]

I have two 1950pros on the way that I plan on useing for a crossfire system or one in each of my systems. I Have not decided yet. If i can remember I will turn off all my other folding for a few days and run just one X1950PRO to get some solid numbers.

[Raptorfury]

What i dont get is why is so hard to work on the G80.
Nvidia claims this GPU is the first to be GPGPU(General Purpose) and also release a language called CUDA - whiche allows developers write a C like code into the GPU.

the only guess i can say is driver support . is why it cant . it also supposedly has the avivo and hdcp as the ati counterpart ...

[Kougar]

Here is your reason:

Once again, we would like to be able to run on Nvidia hardware, but we need driver/compiler fixes from them. You should have a chat with them if you would like G80 support in FAH sooner.

ATI has been taking the time to actually assist F@H with some of their driver releases.

And also what was said regarding nVidia's CUDA:

CUDA is not "just like programming a CPU". It's actually reasonably similar to Brook, but has the added complexity of managing threads and shared memory directly. CUDA is not magic, and for the bulk of the code we use in FAH, it's not going to be much faster, if at all, than the Brook code. The main advantage of CUDA and CTM is that they, in theory, have less overhead than using the graphics APIs. Also, at least for the moment, CUDA requires specific drivers that may impact game play.

Ironically original F@H on the GPU started out for nVidia cards, but the accuracy/performance is not there for G79 and older hardware, so they turned to ATI hardware instead. At the time they also had to "fix" nvidia driver issues to do this.

Source (http://forum.folding-community.org/viewtopic.php?t=16684&postdays=0&postorder=asc&highlight=8800&start=0)

[liqnit]

Here is your reason:



ATI has been taking the time to actually assist F@H with some of their driver releases.

And also what was said regarding nVidia's CUDA:



Ironically original F@H on the GPU started out for nVidia cards, but the accuracy/performance is not there for G79 and older hardware, so they turned to ATI hardware instead. At the time they also had to "fix" nvidia driver issues to do this.

Source (http://forum.folding-community.org/viewtopic.php?t=16684&postdays=0&postorder=asc&highlight=8800&start=0)
thanks
i didnt know that about cuda
i wonder why nvidia is less helping???

[Kougar]

To give nVidia the benefit of the doubt, I'd assume because they have both hands and feet full trying to create Vista drivers... Their old approach of a unified driver set for their cards simply no longer is possible, so they've had to create new drivers for both Vista x32 and Vista x64, for both G79 and G80 hardware.

It's actually still more complicated then this. Ontop of 4 different drivers, they must create a 5th for G80 SLI, then a 6th for G79/older SLI. So that is 6 completely new drivers just for Vista they had to build from scratch, not even counting G80 drivers for XP. ;) nVidia didn't spot the roadblock to their usual driver approach until they smashed head on into it, and they are still playing catchup as obviously nothing more than beta SLI drivers will exist yet for Vista until about March.

ATI seems to have spotted their roadblock early on, and rebuilt their Catalyst driver system to avoid it this time... since they are not playing catch up they've been pretty happy to help Stanford with their folding@home project, since they can indirectly turn a profit via their stream initiative. ;)

[liqnit]

I see now that Nvidia really has a big job on hand.
I hope it will be finished soon and would be stable .
And still i think they should do something in the direction of Folding

[Kougar]

Well, they had way over a year of time to know this was coming, but obviously only ATI put that time to use! I won't give nVidia any slack on that part of the story... Studying the night of the final exam is never a good thing, especially when ya have 6 of them... ;)

[liqnit]

Well, they had way over a year of time to know this was coming, but obviously only ATI put that time to use! I won't give nVidia any slack on that part of the story... Studying the night of the final exam is never a good thing, especially when ya have 6 of them... ;)
agree , but maybe Nvidia is preparing some surprise for us all...

[Kougar]

Will, plenty of leaked info on 8900 series, and notice of a "8800GTX Ultra" in the drivers themselves.... I'd say the cat's out of the bag on any surprises!

[BigD]

gotta love how nvid makes the same mistake over and over and over again with future cards in there drivers

[Kougar]

Actually I think that was ATI... :whistler:

It's ironic how major companies like to play trading places though... ATI will be on the third release of their new drivers by the time they launch DX10 cards, and their current drivers are already doing pretty good.

[Scott Sherman]

I think the drivers for the 8800 cards in XP and Vista have been a mess. They had better start getting the driver side of the house fixed fast. AMD is getting ready to release a product after 13 spins until the got what they wanted out of it, and I bet it is going to kicks NVIDIA's ass all over the place.

The 8800 hardware is great, we just need drivers bad to show it off. At least 2D and 3D quality have improved greatly with the 8800 cards.

[westy87]

In regards to this whole F@H thing, I read the FAQ, yet it isn't all that clear about little things. So, all we have to do is sign up, and when you have it running it uses idle cpu power to run some sort of program that helps out some university lecturers?

[Kougar]

How on earth did you arrive at that conclusion, if I may ask??

The entire purpose behind Folding@Home is to find cures for debilitating diseases such as the various types of cancers, Alzheimer's, Parkinson's, and still others. The project was started because Stanford by far doesn't have the supercomputing reasources required to simulate protein folding (Think gene code in action within your body, it's the software that runs cells and everything else). They turned to distributed computing, where people donate what CPU time they can or are willing to, and through the "donations" of spare CPU cycles Stanford can continue the research to finding these cures... According to their own FAQ, if Stanford had access to use all the supercomputers in the wrold it would still not be adequate for this job, hence why they turned towards the innumerable number of users and partially idle computers in the world instead.

Some data on this: http://www.dailytech.com/article.aspx?newsid=6609 (http://www.dailytech.com/article.aspx?newsid=6609)

[westy87]

Well yes, by university lecturers, I did mean the project they are working on ie Disease research. This part in the faq:
Q: How long do I have for each project?
A: Every project is different in size and complexity, and comes with it's own deadline. Most are a month, but some are two months. Refer to the "unitinfo.txt" file after you configure your client.

What does that mean exactly? Im assuming theres just some kind of configuration settings u have to change each month to suit the area of research they require? I would really like to get into this to help out, but want to know what I have to do first.

[Kougar]

That paragraph is simply saying you can refer to that file to find out each individual WU's deadline. You do not have to do anything with that.

WUs (Work Units) expire after so long so that Stanford can send the same WU to another computer to get it done before it causes a problem. Much of these WU's are based off of prior work done, so if someone downloaded a WU but never completed it and sent it in, then it would eventually cause delays. The program itself will automatically throw out expired WU's and DL a new unit to work on instead.

The GUI/graphical version is pretty much maintenance free, just start it and it will run on it's own, and stop it before playing any games since it uses openGL. Single WU's usually take a day or more to finish, so don't expect instant results with this. :) If you are comfortable with software apps then ya can try the client version, it is a simple DOS window that runs in the background. More detailed setup info is required for the client version, this may be of help. http://fahwiki.net/index.php/How_tos

[westy87]

1 more question i should have asked before. In this thread it was discussed about using the cpu to do the work, or the GPU. Which do you recommend with my hardware? And is there an option to choose when you set it up. I'm asking now because i'm home for hollidays from uni and left my main computer there, so can't set it up yet/

[werty316]

Currently only the including X1600, X1800, and X1900 series GPUs from ATi can be used to fold so your only option is to fold using your CPU.

[westy87]

i downloaded the client but am unsure as to how to set it up and make it work...lol dont laugh. How do i do it, or point me towards FAQ that tells me?

[Kougar]

Here ya go: http://forums.anandtech.com/messageview.aspx?catid=39&threadid=1725668&enterthread=y&arctab=y