Status of GPU Folding

[CoW]8(0)]

Anyone know when someone somewhere will release a folding client for GPU's? Doesn't seem anything new has happened since August

 

[SmokeRngs]

I haven't heard a thing about it. I find that really annoying considering the vid cards they have out now and the power that could be made use of. I personally won't be able to do anything like that with my ancient card, but I won't be using this thing forever.

The new ATI GPUs could possibly change that, though. They were designed for people to use the vid card for things like this. I hope Stanford picks up on it if ATIs architecture does allow for an easy implementation of it.

 

[brycejones]

The new ATI GPUs could possibly change that, though. They were designed for people to use the vid card for things like this. I hope Stanford picks up on it if ATIs architecture does allow for an easy implementation of it.

Maybe crossfire will be good for something after all.

 

[Kvothe]

Here's a read from the guys at Stanford dicussing alot of the in's and out's of GPU's and folding: http://forum.folding-community.org/...=0&postorder=asc&highlight=gpu+client&start=0
and here is a link to some of the results that they've gotten thus far: http://folding.stanford.edu/FAQ-highperformance.html

There's a ton of good reading there, but they're close and it'll be interesting to see how this ends up doing.

 

[CoW]8(0)]

Yup, they seem to be getting close to it but not quite there.

September 2005 Vishal has found that the new changes aren't quite there yet. We're also benchmarking/testing/evaluating new hardware that has come out.

I hope sometime in 05' this'll be ready

 

[brycejones]

That is some interesting news. It would be nice to see my 7800GT cranking out some WUs.

 

[CaptRingold]

Interesting.. F@H was already a slight excuse for a powerful CPU that runs 24-7, now it justifies a good video card too

 

[MN Scout]

This pdf is a nice read. General Purpose Computation on Graphics Processors (GPGPU) by Mike Houston, Stanford University

 

[CaptRingold]

Okay, partly OT but regarding that link to the high performance FAQ

I dont understand whats wrong or so difficult about the SMP project they're working on. I've ran SMP on a Dual Athlon MP box and now an X2 box fine for a long time without a problem. The performance seems okay; both utilize 100% of their cores/processors, so I take that to mean there isn't a large bottleneck from the memory or something. How hard could it be to make a console or GUI client that installs and says "Okay, user has X cores. Downloading additional clients... initializing X gromacs/tinker/etc cores." ?? Or is there more performance to be had by enabling a single gromacs or tinker core to scale very well over multiple cores? I can understand if THAT is the problem, given all the complaining by the gaming industry on multithreaded apps, but just having an app that knows well enough that i can run multiple cores at once isn't too hard -- BOINC for example already does it by default.

 

[brycejones]

I had the impression from the SMP portion of the FAQ that they were trying to figure out how to multi-thread a single work unit so both cores/cpus were working on the same WU.

 

[Kvothe]

If I am reading it properly, it appears that their goal is getting Gromacs to run as a mutithreaded app, so the single instance can run over multiple cores, just from what they've stated, it appears the the biggest problems are that so far, X number multithreaded doesn't equal the individual performance of X number of single instances.

 

[brycejones]

So it would be nice if they initally focused on delievering a core that could detect the number of instances required and work x number of WUs at the same time. At least as a quick solution that allows them to fully leverage the available CPU processing time while they figure out how to multithread the core.

 

[marty9876]

So it would be nice if they initally focused on delievering a core that could detect the number of instances required and work x number of WUs at the same time. At least as a quick solution that allows them to fully leverage the available CPU processing time while they figure out how to multithread the core.

I guess I don't follow this. Can't you just install as many instances as you want with the -local flag? OC Forums have one clicks for X number on instances.

 

[CoW]8(0)]

Finally, time to make some good use of the GPU.


Maybe this thread should be stickied to make sure all the folding potential is well informed.

 

[lozaning]

as if my case didnt run hot enough with my oc'ed a64, now i get my x1800xl pumping out another 80 degress runnign at 100%, sounds like a furnace

 

[SmokeRngs]

as if my case didnt run hot enough with my oc'ed a64, now i get my x1800xl pumping out another 80 degress runnign at 100%, sounds like a furnace

Yes, it can act as a furnace. It's great for heating your house for the winter. That way you don't have to worry about turning your actual furnace on. It saves power since you were going to have the system on anyway.

 

[Kvothe]

Yes, it can act as a furnace. It's great for heating your house for the winter. That way you don't have to worry about turning your actual furnace on. It saves power since you were going to have the system on anyway.

I can see it already, people world wide are going to be going out, buying boxen and box fans, and replacing their heaters with OC'd folding boxen.

 

[SmokeRngs]

I can see it already, people world wide are going to be going out, buying boxen and box fans, and replacing their heaters with OC'd folding boxen.

Yes, it's a dream I have. I'm not sure if it will be accomplished in my lifetime, but I can keep hoping.

 

[Oldie]

Yes, it can act as a furnace. It's great for heating your house for the winter. That way you don't have to worry about turning your actual furnace on. It saves power since you were going to have the system on anyway.

Heh,

This is actually how I heat my workroom. In the winter I close the vent, and it usually stays a nice 70-72F inside. REALLY!

 

[BillR]

CPUs are designed from the ground up to run at 100% full power 100% of the time and then we go and over clock them to well beyond all reasonable expectations and for some reason they keep on giving us that good long life we have come to expect.

A GPU on the other hand rarely runs at 100% and when and if it does it doesn’t do so for sustained periods even with the all the toys turned on in games.

My concern is these things already run very very hot as it is, even when water-cooled. Somehow I doubt ATI or nVidia have taken into account 100% GPU usage on a 24/7 basis doing chores for which they were never intended and at the same time providing us with useable video.

As video companies their main thrust is to give us high FPS and good color etc at reasonable (save of course the latest greatest bleeding edge card) price.

I’m sure there are behind the scenes talks between the people at Stanford and ATI and nVidia but I find it hard to believe there would be enough demand for such a card to justify the overbuilding and redesigning necessary to sustain both video and additional FP calculations and still give you some kind of reasonable warrantee at any kind of reasonable price.

Just some thoughts

 

[Duster]

Just some thoughts

some good points

I would like to see multi core stuff.

 

[SmokeRngs]

CPUs are designed from the ground up to run at 100% full power 100% of the time and then we go and over clock them to well beyond all reasonable expectations and for some reason they keep on giving us that good long life we have come to expect.

A GPU on the other hand rarely runs at 100% and when and if it does it doesn’t do so for sustained periods even with the all the toys turned on in games.

My concern is these things already run very very hot as it is, even when water-cooled. Somehow I doubt ATI or nVidia have taken into account 100% GPU usage on a 24/7 basis doing chores for which they were never intended and at the same time providing us with useable video.

As video companies their main thrust is to give us high FPS and good color etc at reasonable (save of course the latest greatest bleeding edge card) price.

I’m sure there are behind the scenes talks between the people at Stanford and ATI and nVidia but I find it hard to believe there would be enough demand for such a card to justify the overbuilding and redesigning necessary to sustain both video and additional FP calculations and still give you some kind of reasonable warrantee at any kind of reasonable price.

Just some thoughts

To a certain extent I would agree with you except for maybe the newest ATI GPU's. They are actually designed to be used for something other than just processing some video.

Also, they would actually have to qualify the parts at a certain temp otherwise they would fail all the time when being run hard. People can and do play games for extended periods of time. While they may have not been designed for 24/7/365 they would have to have some type of heat stressing to make sure they would take the stress that was thrown at them to begin with.

Also, you may want to take into account that not every part of the GPU would be used while doing this probably. That would drop the temps a bit while in use.

Basically, we really don't know how it would affect it without some type of testing done. It would be interesting to find out the results. We may be finding the aftermarket heatsink business for GPUs suddenly growing soon, though.

 

[Oldie]

Actually we do know that Vista will be rather GPU dependant for quite a bit of the shell effects. Perhaps we're seeing the new ATI chips as the first graphic cards designed with this in mind.

 

[DarkBahamut]

My concern is these things already run very very hot as it is, even when water-cooled. Somehow I doubt ATI or nVidia have taken into account 100% GPU usage on a 24/7 basis doing chores for which they were never intended and at the same time providing us with useable video.

This was my main concern also. I honestly dont believe a video card running at 80C (which most high-end cards do on stock cooling) is going to have a very long life span at 100% load 24/7. Even with an after market aircool these days, your still looking at 60-65C under load, which helps alot, but its still fairly hot, all things considered. Maybe if we could keep these cards to 50C or less under load then we'd be getting somewhere, but that requires water cooling at least these days, which isnt an option for most people.

I dont think theres really any worry of our videos cards dying in a month or anything like that, but without temps below 50C, i doubt they will be lasting as long as CPUs either. However, as GPUs do become more designed for things other than just 3D graphics they will probably hold up better under 100% load 24/7.

The after market video card cooling market will sure boom the second we can use GPUs like CPUs though. Im just picturing the 120mm fans stuck on G70s and R520s already .

 

[[Spectre]]

Bump itty bump

http://techreport.com/onearticle.x/8887

 

[CoW]8(0)]

From this info, would it even be worth using anything other than a X1000 series to fold? Considering that the even the 7800GTX doesn't do very well at GROMACS.

 

[brycejones]

Very intresting article. Now imangine 4 X1800s running on that new gigabyte socket 775 board being previewed over at anandtech that has 4 16x pci-e slots. http://www.anandtech.com/mb/showdoc.aspx?i=2561

 

[CoW]8(0)]

*Bump*

October 2005 With the release of the R520 card, it looks like we're getting close. Our unoptimized code is already looking good and we are working to optimize it.

Let's hope nVidia cards are optimized a bit more

http://folding.stanford.edu/FAQ-highperformance.html

 

[unhappy_mage]

1) Hahaha they compare it to an Intel... Compare to a real processor you wimps
2) Stop posting interesting links, I'm trying to get my homework done

3) Okay, done reading. Looks interesting, can't wait for release! Of course, this means I'll have to buy a real graphics card

 

[Mattman]

The one thing I couldn't get from the PDF on the latest X1800 developments was whether or not the lower-end ATI cards (same basic core as the X1800) will be greatly affected by lower clock speeds, fewer pipelines, etc. I still don't think I would shell out the big bucks for a high-end video card even if it was a fast folder, but if a sub $200 card was nearly as fast as the big boys, it would definitely influence my decision about my next graphics card in the continuing ATI vs. Nvidia battle.

This is definitely exciting for the home users and home farmers since they're more likely to have the high-end hardware to take advantage of this newfound power.

 

[MN Scout]

It'd be nice if Ati 9800s were able to help. I know that they wouldn't put out huge numbers but it would be an added something to what ever the main processor is putting out. I have a 9800 so that's an added bonus. I know that they started this GPU project back then so I"m curious if any AGP cards will be supported or if it'll just be pci-express cards?

 

[CoW]8(0)]

It looks like a client will be out within the next month or so. I'm really hoping the nVidia series (6 series to be specific) will be able to contribute

 

[SpoogeMonkey]

That's good news. I'd really like to put my sli'd 6600gts to use. And the 7800gtx I'm buying myself for xmas (unless something better comes out in that time frame)

As far as the dual core situation: although I don't have an x2 yet, I like the idea of running one instance per core. Even if you're encoding video or doing a typical 18hr WoW session, you still have 1 core folding 100%, while the other core takes care of your pr0n.....errr......gaming habits in the background.

 

[CoW]8(0)]

Update:

December 2005 We are working on optimizing the newest code. We are seeking help from GPU manufactures to tune our code on their hardware.

Hmmm, not much detail on real progress, hope everything is going smoothly there....