PPD on GPU-based folding, ATI vs Nvidia?

Emission

Emission

Supreme [H]ardness
Joined
Dec 6, 2005
Messages
4,420
This is an inquiry I'm throwing out there to all GPU folders. I once used Nvidia cards for folding, but have since sold/got rid of them for various reasons (ex: my last one died), and have switched over to ATI for budget reasons. With my E-VGA 9600 GSO, I was netting around 4800 PPD, GPU alone. When I moved over to an E-VGA 8800 GTS (G92), That increased my PPD to around 6000 or so. That card faced an accidental death, and I have since upgraded to a XFX HD 4850. Now my PPD is just under 1200.

Can someone explain the difference in software/hardware between the ATI and the Nvidia GPUs and why the Nvidia's are netting such high scores? I know that the HD 4850 is a small upgrade from the 8800 GTS (G92) in terms of graphics capabilities, but I was also pretty sure that the ATI card has a stronger processor overall, so what's going on here that's limiting the points on the ATIs? What's the difference?
 
the GPU client seems to favor CUDA. there is a new cleint in the works and many of us are hoping for stronger numbers from ATI GPU's, but at present CUDA is the way to go for GPU folding.

just looking/comparing the specs would lead you to beleive the ATI cards would be big producers, but it just aint so....hopefully we will see some inprovement in the upcoming GPU3 client, and openCL.

i know there are some others here who can really lay down the facts on this, i am simply speaking from my experience. my 3850 cannot/will not come close to my 9600GT let alone my 9600GSO or 260.
 
All your assumptions are right.

However, The folding client and the WU that we are getting are made for a smaller number of very fast shaders. ATI has a large amount of slower shaders. The WU/client just doesn't know how to scale over them.

Also from my understanding the NV cards do each calculation once, where as the ATI (because of the way they are designed) have to do the same calculation twice.

With the current software we have, the deck is stacked against the ATI cards.

1200ppd is very low for a 4850 however and I hope someone that folds with one might be able to help you out.

There is hope. We are getting a new core (in a few months?) that should use the ATI cards more to their potential.

 
Thanks guys.

@ Kendrak: I figured it was an elementary hardware difference that made this happen.

Hopefully next time around I can get a better deal on Nvidia GPUs, or put together a spare machine specifically for folding.
 
Anytime you need help on configs or tracking down a good deal, just ask.

I'm always hunting down cheap folding hardware.
 
Kendrak said:
Also from my understanding the NV cards do each calculation once, where as the ATI (because of the way they are designed) have to do the same calculation twice.
Click to expand...
That's because of the way the software is designed, not because of the hardware. ATI cards since the 4000 series have had the ability to avoid repeating calcuations by storing results in the LDS, but that functionality was never added to the ATI GPU client, so they're all running in R600 mode. Hopefully GPU3 will deal with that, but I'm not getting my hopes up.
 
Zero82z said:
That's because of the way the software is designed, not because of the hardware. ATI cards since the 4000 series have had the ability to avoid repeating calcuations by storing results in the LDS, but that functionality was never added to the ATI GPU client, so they're all running in R600 mode. Hopefully GPU3 will deal with that, but I'm not getting my hopes up.
Click to expand...

see, i was right...someone here knew the facts :p

it would seem that there could/should be a switch in a command line to disable or force a different mode. although i only have one ATI forlding, i would certainly like to get more perts outta it. its more than a little disappointing when compairing my 9600 and 3850 as far as production goes.

here is hoping for some GPU3 love for the ATI.
 
xxGriff said:
it would seem that there could/should be a switch in a command line to disable or force a different mode. although i only have one ATI forlding, i would certainly like to get more perts outta it. its more than a little disappointing when compairing my 9600 and 3850 as far as production goes.

here is hoping for some GPU3 love for the ATI.
Click to expand...
GPU3 probably won't help 2000 and 3000 series cards. If it does have any improvements, they'll most likely apply only to 4000 and 5000 GPUs because the older GPUs don't have the LDS.
 
xxGriff said:
see, i was right...someone here knew the facts :p

it would seem that there could/should be a switch in a command line to disable or force a different mode.
Click to expand...

For that to be an option, it would have to of been coded.... never got coded.
 
1200 ppd does seem low. I get about 3k ppd on my 4890s and the 4850 is clocked 71% as fast as the 4890s. Are you sure it is getting out of 2D mode? pull up GPU-Z or Overdrive console or something and see what frequency your GPU is running at. Mine drops to 240 Mhz in 2D mode. 625 Mhz is full power.
 
sorry, i should have been more clear, i understand it must be coded, i was only saying i think it should have been there.

i did not expect any real improvements to 2000 or 3000, i understand the architecture is simply not there. having said that...i do think and hope that the coding for 4000 and newer would include some refinement to take advantage of the strengths in ATI's silicon.

ATI GPU's are quite capable of handling some impressive number crunching, its just not there under GPU2 comparative to nVidia GPU's. having a 790 AM3 crossfire board makes the 5000 series tempting on a purely gaming point of view, but i dont game much like i used to.

btw i also concur, 1200 is very low.
 
I checked the stats again just now and it seems a little more accurate now. It's something closer to 2000 PPD as it stands right now. I had a feeling my PPD was low, but it might not have been reporting it properly on my [H] Folding page which is another issue entirely. I have it clocked at 725 Core / 1100
Memory, and that is confirmed by RivaTuner and the fact that my times noticeably improve with the scaling up of the speeds.

Looks like the GPU3 client should be something to look forward to then?
 
the PPD you see in your monitoring program and what you see on the website are 2 different numbers.. the numbers on the program are based off if you ran that same work unit type for 24 hours where as on the website its your average over the entire 24 hour period.. since there are different types of WU's your PPD will very according to the type of WU you are running.. so your PPD in the monitoring program will change depending on that WU..
 
leave it to zero lol, that guy knows everything!

Anyways, only other thing that seems to have been left out is that the points system (I believe) for ATI cards is still gauged off an old 2 series card isnt it?
 
jeremyshaw said:
Is that benchmark for all cards? I cannot find the nvidia benchmark.
Click to expand...

All cards.

That is one of the main reasons NV cards get so much more ppd. They finish the WU faster than the ATI cards and thus get a big bonus.

If they are able to speed up ATI cards, expect a drop in NV ppd in step.
 
roger, 3 series, I knew it was before the 4 series, just couldnt remember which one
 
You must log in or register to reply here.
Back
Top