How To Build The Best Folding Rig

[PoppaGeek]

How To Build The Best Folding Rig

[570091D]

the chart with ppd for video cards is what i've been looking for, it's interesting that the gts250 and gtx260 are almost equal in production. i was under the assumption that the gtx260 would be faster due to the more efficient micro-architecture.

[PoppaGeek]

I saw that and am not sure that is real accurate. The 250 has 128 shaders and the 260 has either 192 or 216. The 260 does produce more points in GPUGrid and other charts I have seen shows it producing more. But with so many different size WUs I think it is hard to really pin down a set number for PPD.

[OldChap]

The general rule for me with a 98gtx vs. a 260 (192) is that an average of 500-800ppd extra is seen on the 260 when seen as a daily output.

[DR3]

the chart with ppd for video cards is what i've been looking for, it's interesting that the gts250 and gtx260 are almost equal in production. i was under the assumption that the gtx260 would be faster due to the more efficient micro-architecture.

The both GTX260's have better effective performance because the GTS line has a glitch with the folding clients and hangs up occasionally.
The GTX260/216's I have in my folding farm have been rock solid.

But back to the topic.
The best folding rig would be the mobo with the most PCIEx16 slots, that currently is the ASUS P6T7 Supercomputer with 7 PCIEx16 slots.

The highest performing video card currently is the GTX295.
To get seven of these to fit you would need to get the single pcd version and liquid cool them.

In addition you would need two power supplies to feed the beast.

And lastly the rig would have to run linux to accommodate the 14 GPU's.

Most folders would not go through this much turmoil just for a single folding rig. They would instead go for two rigs air cool them. It cost about the same and would be far less turmoil.

[PoppaGeek]

Another GPU chart for F@H at TechPowerUp

[570091D]

thanks for the heads-up on 260 performance, i just checked the link in the milestones thread, it's very tempting to pull the trigger on a couple of those! however, i'm also tempted to just wait until dx11 hardware hits the shelvs, then i'm sure we'll be grabbing 260's for less than $100.

the rule of one cpu core per two gpus seems a bit off to me. when i set the affinity for my f@h clients to one cpu core, there is still hardly a load placed on the cpu (watching with windows perf monitor). but this could be attributed to the overclock on my cpu....

[INFRNL]

I know that first chart is wrong; there is no GTX 260 that outfolds a GTX 275.

I think I might have to get a gtx 260/216 and compare it to my GTX 275's, cause what the members are stating their ppd is not that of a GTX 275.

good looking out though PG

[PoppaGeek]

I know that first chart is wrong; there is no GTX 260 that outfolds a GTX 275.

I think I might have to get a gtx 260/216 and compare it to my GTX 275's, cause what the members are stating their ppd is not that of a GTX 275.

good looking out though PG

There are several problems with that chart. They have the 4850 at 2063 ppd and mine did 3000+ ppd. They have the 260 (192) below the 9800+ and GTS 250. I fear they did not use the same WU to bench or they simply screwed up the chart. TechPowerUps chart appears much more realistic.

[570091D]

hrm, the tpu chart puts the 9800gtx at 5900ppd, but i have a pair of 9800gx2s and i've never seen those numbers from any of my cores. with 353 wu's i'll get 5200ppd max, with 511 wu's they drop to 3500ppd.

[Sparky]

Maybe it is based off the old units that scored higher? The beta test units that were making rounds for the first several months when the client was released.. But then, with GTX260 in there, that wouldn't make sense.

[DR3]

I know that first chart is wrong; there is no GTX 260 that outfolds a GTX 275.

I think I might have to get a gtx 260/216 and compare it to my GTX 275's, cause what the members are stating their ppd is not that of a GTX 275.

good looking out though PG

I agree, something does not smell right with that chart.
Wonder if they just made it from user input or actual benched the cards?

A GTX 260 216 or not will not out fold/perform a GTX 275.

[INFRNL]

I think we need to come up with our own chart of some sort. The one chart link from actual users is out of date. I have nopt been keeping track in a while, but I started a chart which has some of my ppd from my 9800GT, GTS 250, GTX 260/192, and my GTX 275's.

the gts 250 and gtx 260/192 are very close together; I think just a few hundred ppd difference in most cases. I believe the 260/216 is somewhere in between a gtx 260/192 and a GTX 275. I think my 275's still produce 1-1.5k more ppd than a gtx 260/216 if not closer to 2k on some wu's.

I need updated info from the 260/216 owners to compare. or just need to get a 260/216 to compare myself and see if the cost difference is worth it for a GTX 275 over a 260/216. IMO the 45-$50 is worth it to get the GTX 275.

just my 2 cents

Bottom line is that we should come up with our own chart for more accurate comparisons for our members

[coo-coo-clocker]

Agreed, infrnl.
I also would like to see empirical data on how these cards PPD scales relative to shader speeds, and their power consumption. A great chart would show:

Card | Shader Clock | Watts | PPD | PPD/Watt | WU

[DR3]

Agreed, infrnl.
I also would like to see empirical data on how these cards PPD scales relative to shader speeds, and their power consumption. A great chart would show:

Card | Shader Clock | Watts | PPD | PPD/Watt | WU

That would be a great chart for folders to have.
Would also be nice to know if shader clock speed and ppd is linear or not.
I have a hunch it is not linear.
If this is the case then there would be a sweet spot for each card.

Is there a reliable and fairly quick benchmark test for shader stability?

[INFRNL]

so who's the smart guy that wants to get it started? any takers...PG? I nominate PG if he's interested as he is always bringing us a lot of useful info.

I think we could start a thread with Ideas and could possibly start posting our specs, settings and ppd based on the different wu's. Of course we would have to go off some standard setting.

Maybe we could add it to the ppd thread, unless it would be a better idea to start a new one with just our data.

I think something like this would be nice for the members and could potentially eliminate a lot of re-asked questions; which I am guilty of myself.

[PoppaGeek]

I think it is a good idea and a new project would be good for me. I can start it as a Google Docs spreadsheet that anyone can see and is easy to change as ideas for changes come along. When it gets more or less stable we can try some embedded chart in a permanent thread. Sound reasonable?

I'll start a new thread for this project.

[INFRNL]

Thanks PG; I knew you would be a good candidate for this project. Willmove over there and see what we can help with

[DrJay]

The both GTX260's have better effective performance because the GTS line has a glitch with the folding clients and hangs up occasionally.

DR3,
Could you clarify what you mean by "hangs up"?
I'm getting ready to add a GTS250 to one of my folding boxes....this has me a little concerned.

As to the ppd chart in the article, I agree there seem to be many discrepancies. However, it seems they are showing stock clock frequencies. Wouldn't 192 shaders at 12xx Mhz be pretty close to 128 shaders at 18xx Mhz. I think some of the manufacturer OC'd models are at 1890 Mhz.

[DR3]

DR3,
Could you clarify what you mean by "hangs up"?
I'm getting ready to add a GTS250 to one of my folding boxes....this has me a little concerned.

As to the ppd chart in the article, I agree there seem to be many discrepancies. However, it seems they are showing stock clock frequencies. Wouldn't 192 shaders at 12xx Mhz be pretty close to 128 shaders at 18xx Mhz. I think some of the manufacturer OC'd models are at 1890 Mhz.

I do not own a GT250, but it is my understanding that the GT200 series requires a little more babysitting.
This was highlighted in the GPU Farm thread.
That is why I went with GTX260 instead of the GTS250s for my gpu farms.

the gts 250's do about 4k-6k at shader 1890.. you'll rarely see them all on 6k units at the same time.. on average mine did about 5k each when I had all 4 of them on fah.

they do about 9-10k each on gpugrid

also, they have a bad habit of dropping into idle 2d clocks after periodic faulty wu eue's, which do happen occasionally even on a 100% stable card... and the only way to kick it back into full 3d clocks gear is a reboot. so that may be problematic on a big farm with a bunch of gts250's. the 9800gtx+ cards did the same thing. I dont think that issue has ever been resolved. but my 4 have not exhibited that behavior since I moved them to gpugrid, so it has to be something to do with the stanford client and that particular chip (9800/250).. pande wont own up to it though, I've asked on their forums too many times and got the brush off "deal with it" answer everytime.. they keep telling me it's my hardware. and to ask nvidia. I asked nvidia and the only thing they could tell me was to try riva tuner force 3 d clocks. it didn't resolve the behavior and evga would not accept a rma based on that. so go figure, all I can say is based on my experience with fah and gts250 chips..

Unitl fah or nvidia driver comes up with a solution to that "drop into idle mode in while folding" bug, I would consider a different card for folding. they are a nice card for the money, plus you only need one pcie power for each card. but hat little bug is a headache if you like your clients to be left in non-baby-sit mode..

So maybe it has been fixed.
Let us know what you get.

[DrJay]

Thanks for posting all that, DR3.

I suppose I'll need to get an eye on it for a while after i get it set up.

[INFRNL]

I never had a problem with my GTS 250's, but I only had them for a month or so before trading up

[mike047]

Re 250's

I never had any issues running a single card. When running 3 on a Evga 780i, periodically one would drop to half speed. Reboot rectified the issue.

[[XC] riptide]

That would be a great chart for folders to have.
Would also be nice to know if shader clock speed and ppd is linear or not.
I have a hunch it is not linear.
If this is the case then there would be a sweet spot for each card.

Is there a reliable and fairly quick benchmark test for shader stability?

Well IIRC Oldchap showed that Shader clock can be roughly linear BUT GPU Core clock has a non trivial effect aswell. I remember I had to eat crow at the time.

As for graphs there's already a GPU2 PPD thread, http://www.xtremesystems.org/forums/...=191674&page=9 , and at the top of the forum theres a PPD rate thread also.

[OldChap]

The info is old now but was done when all work used the same core and all results had the same value. I don't remember the thread this was posted in but the original is at: http://forums.pureoverclock.com/showthread.php?t=4258

The crow was cooked tho'