The Folding@Home GPU Thread

[GLeeM]

"the multi-GPU client will not work with cards configured for crossfire. Crossfire couples the cards in a way that works for graphics, but not for computation."

From here: http://forum.folding-community.org/ftopic16669.html

 

[Kinslayer777]

As you may know, I have a Saphire X1900XT 256mb. I noticed someone once oft handily wrote that is fan wasnt running because there was some sort of problem with the F@h instance not working. This made me realize: When i'm just regularly doing light tasks, say, using openoffice, my little loud annoying card fan runs and lightly blows hot air. When i'm running F@H GPU it runs loudly and CONSIDERABLY hot air comes out. Is this normal? It's just a little concern that made me wonder if my card is perhaps malfunctioning for being so hot. It's not overclocked or anything either.

 

[CupCak3]

nope all is well. i run mine 24/7 with no problems at all

I have the same card as you btw

 

[Avalon]

Finally got my X1900AIW working great, and I'm using the 6.5 CATs. Seems to take about 20-21 hours to finish a WU, and I get 330 points for it. Not that bad considering it took 3 days to complete a 500 point WU with the CPU client on my Opty 144. Seems I'm only getting 2x the amount of work done per hour compared to using the CPU client. I thought it was supposed to be 10-20 times faster?

 

[GLeeM]

The science gets done faster.

They benchmark on current hardware and as new faster hardware comes out that will get better points per day because the benchmark will still be the same.

They are trying to base the points on the regular clients' points, but give a bonus for more hardware used.

 

[Duvie]

Originally posted by: Avalon
Finally got my X1900AIW working great, and I'm using the 6.5 CATs. Seems to take about 20-21 hours to finish a WU, and I get 330 points for it. Not that bad considering it took 3 days to complete a 500 point WU with the CPU client on my Opty 144. Seems I'm only getting 2x the amount of work done per hour compared to using the CPU client. I thought it was supposed to be 10-20 times faster?

For me that would suck considering 600point WU's take me as little as 1day or 25hours with my C2D E6600 at 3.26ghz...I get some of those stinker 2.124,2125, and 2126 and it takes about 36hours for 400 pts...

I think the scoring system sucks at F@H and they should pull their supposedly brainaic heads from their assests...

Either way I am not hearing anyone getting the pipedream numbers the op states above....12hours per WU? In theory it shouldn't make a difference on the cpu if all it is doing is feeding the gpu and it is crucnhing...I hear ppl getting 17-21 hours from a wide array of cpus...I dont see any 12 hour completions as average....

I wouldn't even waste my time having the gpu run one and tie up a C2D core running at 3.4ghz on my E6300...i think it does better...

So unless F@H is going to wake up and score them appropriately to the Bullshite of them telling us it was 10x faster, it is a collosal waste for many of us....
I give this whole thing a resulting thumbs down.

considering how much this thread has been updated I see that view is shared by many....

 

[CupCak3]

Originally posted by: Duvie

For me that would suck considering 600point WU's take me as little as 1day or 25hours with my C2D E6600 at 3.26ghz...I get some of those stinker 2.124,2125, and 2126 and it takes about 36hours for 400 pts...

I think the scoring system sucks at F@H and they should pull their supposedly brainaic heads from their assests...

Some people fold for the science, some for the competition and some for both. If you have such a problem with the current scoring system, post your concerns here.

Originally posted by: Duvie
Either way I am not hearing anyone getting the pipedream numbers the op states above....12hours per WU? In theory it shouldn't make a difference on the cpu if all it is doing is feeding the gpu and it is crucnhing...I hear ppl getting 17-21 hours from a wide array of cpus...I dont see any 12 hour completions as average....

I wouldn't even waste my time having the gpu run one and tie up a C2D core running at 3.4ghz on my E6300...i think it does better...

pipe dream? the score of ~600 PPD is based on 512meg x1900xt. With increased speeds my 256 meg card is putting out these numbers. I"m sure if avalon OCed his card, he would get similar numbers. OCing a 512meg card will net 800+ PPD.

And your 600 PPD WUs will be drying per the Pande Group. Good luck getting 600+ PPD out of both your cores combined once that happens.

Originally posted by: Duvie
So unless F@H is going to wake up and score them appropriately to the Bullshite of them telling us it was 10x faster, it is a collosal waste for many of us....
I give this whole thing a resulting thumbs down.

considering how much this thread has been updated I see that view is shared by many....

view shared by many? yours is a bit extreme from the posts I read. As of now, CPUs can do work that GPUs cannot, thus carrying a small point premium. Giving huge point bonuses to GPUs right now would begin to "alienate" CPU users. (at least in the eyes of the Pande Group) Points values are still under consideration, so your feedback to the Pande Group is valuable at this time.

 

[Markfw]

I have a 256 meg X1900XT, and it takes me 19 hours on an X2, 3800@2.3. It has one CPU just for the GPU, and one doing F@H. You mean to tell me that if I had 512 meg instead of 256 meg card, I could do that in 12 hours ? And it won;t run faster than 500 mhz. I tired all of the suggestions, and I don't know how they do it, but mine will only run @2d speed.

 

[GLeeM]

Originally posted by: Duvie
For me that would suck considering 600point WU's take me as little as 1day or 25hours with my C2D E6600 at 3.26ghz...I get some of those stinker 2.124,2125, and 2126 and it takes about 36hours for 400 pts...

PandeGroup gives bonus points for WUs that use more computer resources.

The 600 pointers are some of those bonus WUs and get 2X the points. If they did not get a bonus they would only be 300 points and the 2124-2126 WUs at 400 points would be better.

There has been some talk about doing away with bonuses because of all the misunderstandings. Which would be too bad, I thought the idea of giving more points for more resources used was a fair way to recompense those will to give those resources.

As for not being able to get bonus WUs all the time - PandeGroup has to put what areas of research they need to do in the forefront. Those areas are not always in the areas that the bonuses are.

 

[CupCak3]

Yes, unfortunately how the WU's fold, 512meg cards beat the 256 cards with the current WUs. This is quite a sore spot with me and its something I really hope they fix what I consider a bug. I'd buy 4 256 meg agp 1950pros when they're available if they fix the performance difference but if not it may only be one or two.

Mark: It sounds like you may have not have properly killed your ATI service.


My Setup:

x2 3800+ @ 2.7
x1900xt 256meg 680/783

 

[Kinslayer777]

Originally posted by: CupCak3
The dual core client for those w/ dual pci-e 8x or 16x setups
Download
Configuration

Woot!

I'll look into setting that up asap


Edit: Btw who turned cupcake into a *fish*?!

 

[natethegreat]

Found some news:

G80 will fold also!

enjoy!

 

[sygyzy]

Some questions -

1. How do you know what core or GPU is being used for a particular folding client? For example, I can put two copies in different directories and configure one as CPU #1 and the other as CPU #2 (because I have two cores). When I run them, I just assume they are running independently, but there is no way to verify. Additional, with the GPU client, I call it CPU #3 but how do I know it's using the GPU and not a CPU?
2. Why can't I run 3 clients? The OP said only 1 cpu and 1 gpu even if you have two cores.
3. What do I set for "Memory, in MB, to indicate <2046 available>"? I am asking for each core (CPU and GPU). For example, if I am just running one per CPU core, do I split the memory (1024 each)? What about GPU? I think my card has 512 graphic memory, so do I put that?

 

[CupCak3]

Originally posted by: natethegreat
Found some news:

G80 will fold also!

enjoy!

that is good news nvidia would be stupid to not put themselves in the tech of gpgpu.

now hopefully in the coming month(s) it will take down some of the other card prices

 

[CupCak3]

Originally posted by: sygyzy
Some questions -

1. How do you know what core or GPU is being used for a particular folding client? For example, I can put two copies in different directories and configure one as CPU #1 and the other as CPU #2 (because I have two cores). When I run them, I just assume they are running independently, but there is no way to verify. Additional, with the GPU client, I call it CPU #3 but how do I know it's using the GPU and not a CPU?
2. Why can't I run 3 clients? The OP said only 1 cpu and 1 gpu even if you have two cores.
3. What do I set for "Memory, in MB, to indicate <2046 available>"? I am asking for each core (CPU and GPU). For example, if I am just running one per CPU core, do I split the memory (1024 each)? What about GPU? I think my card has 512 graphic memory, so do I put that?

1 & 2) As of now, the GPU needs a full CPU core to poll and do its dx9 thing. The highest PPD (points per day) and science acheived is with either two GPUs or one GPU and one CPU client running. A 3rd client running on the CPU massively bogs down the machine giving you massive inefficiency.

3) memory is RAM memory from your motherboard

 

[CupCak3]

Originally posted by: Kinslayer777

Originally posted by: CupCak3
The dual core client for those w/ dual pci-e 8x or 16x setups
Download
Configuration

Woot!

I'll look into setting that up asap


Edit: Btw who turned cupcake into a *fish*?!

mooooo

 

[sygyzy]

Originally posted by: CupCak3

3) memory is RAM memory from your motherboard

So if I am running 1/2 cores and a GPU, for both configurations, I put 2048? I would imagine you'd want to split it.

 

[GLeeM]

Some CPU WUs can use lots of ram. The 600 pointers peak at about 120MB.

If you find that F@H is using more ram than you want it to use, you can use that question to tell the servers to only send you WUs that use less ram or the amount you are willing to give.

You can just use the default value (which should be the amount in the system) unless you want F@H to use less ram.

 

[Kinslayer777]

Alas i had hoped my comment was not noticed...
I had misread it and thought that you had posted a new update...

Sigghhh...

But hey

I only have one dreadful qualm with the GPU client. My gpu is dam LOUD and sometimes I have to replace it with the CPU2 client overnight (I keep my precious comp about 8ish feet from my head while sleeping ehehee...)

I know there's aftermarket cooling for the card (or so i think?) but is there anything else i could do? And if i do buy one, would i have any reuse for the current fan? Perhaps resell it for a few bucks so it seems less wasteful to get extra cooling. The noise is the problem not cooling (i dont think) so it doesnt seem justifiable to get extra cooling. However of course extra cooling could run slower and quieter if it were much better. I dont OC.

And in the business of asking questions...Do you think a very light prop next to my gpu vent would be bad for my gpu? perhaps not fully allowing the air out. I think it would be **incredibly** neat to rig a small generator-prop if i can find a light bulb that can run on it. It'd be the F@H light...If i managed to hook it up so onlythe GPU client fanspeeds would make it spin enough....Aww this'd be brilliant . I may have asked it though i forget- i never got to it anyway


Edit: Oh yay! You're back! ...Fishyies smell funny... :\

 

[Duvie]

nvm....

one card has 256 and the other has 512mb of ram...seems to make huge difff....




Does anyone know if cat 6.10's see a huge raise over 6.5's

 

[CupCak3]

some see a bit of a diff... i'm running the 6.10 betas right now. If i were to upgrade my drivers now, I'd switch to the 6.11's which came out a couple weeks ago.

 

[Duvie]

bump...keep to main page...

At 675/800 with my XTX I am seeing 5:35 frames and now avergaing 860ppd off of my gpu...Not another XT for my 3rd box...i doubt it will OPC as far but at stock I was seeing about 750ppd...

 

[CupCak3]

up to the top so everyone can update!

 

[Ayah]

Will somebody buy me a couple X1900s with 512MB of memory?

 

[Kinslayer777]

Ooh, this explains why it stopped working....Thanks!