Maximum PC

Due to a GPU death, I've upgraded a 560Ti to a 660(non Ti version). Everything about the 660 card is bigger, better, faster when compared to the 560Ti. More than twice the number of cores, double the memory, higher clocks all around, and higher memory bandwidth.

So, can anyone explain to me why this "upgrade" is actually running as a downgrade in PPD?

I have compared benchmarks on three different WU types, and this 660 is consistantly 3k PPD lower than the two 560Ti cards still in my network.

Is there some kind of setting I need to change in the config to help this thing work to it's potential?

Folding at Home has not been optimized for Kepler, though they (Stanford) are starting to work towards that end. Until they have a client with full Kepler optimizations, you won't get the full potential from any Kepler cards.

pretty sad really...both my gts 450, and 660ti are working on the same WU8074. it says the gts 450 is averaging is 26149ppd with a time per frame of 2 mins 07 secs, and that the 660ti is averaging 10204ppd with a time per frame of 5 mins 27 secs

kinda messed up that a $300 higher end video card is getting its a$$ handed to it by a card 2 generations old that was a LOW end card to begin with..

@mk96 - what driver version, F@H version, & GPU client version you running???

I just got this Zotac 660 in for a client build, & if eVGA PrecisionX is right, the card has a lot of untapped potential. Maxed it out in Precision @ 1215MHz on the core (Precision won't go any further, but I believe the card will), it's showing 56C temp @ 78F ambient, & only hitting 84% of maximum TDP while hitting 22,700ppd on a p8074. On my GTX-460 pushed all the way up to 825 / 1650 / 2100 (reference clocks are only 675 / 1350 / 1800), I was seeing 17,600ppd for p8074. For Project 7623 (the worst of the 762x projects) the Zotac @ 1100MHz gets 28,300ppd while the 460 gets just 21,860ppd.

Running 306.97 WHQL drivers, F@H program version 7.2.9, & GPU client version 2.25 on both the above mentioned GPU's.

FAH control version 7.2.9
310.70 nvidia driver
no idea on the gpu client version. says FAH core 0x15

perhaps theres flags im missing or somehting, IDK been out of the folding loop for a while back when the version 7 client first was in the works. decided to fire the rig back up for the race to 100

Client version is in the log file:
04:55:00:WU01:FS00:FahCore 0x15 started
04:55:01:WU01:FS00:0x15:
04:55:01:WU01:FS00:0x15:*------------------------------*
04:55:01:WU01:FS00:0x15:Folding@Home GPU Core
04:55:01:WU01:FS00:0x15:Version 2.25 (Wed May 9 17:03:01 EDT 2012)

For a 600 series card that's what you want to see 2.25 since it's the first version that has any sort of coding for Kepler. If that's what you're running (& you should be), you might try rolling back to 306.97 WHQL drivers. From the 310 series release notes:
It could be that with 600 series cards, these power optimizing enhacements don't work out so well for folding...

I have:
However, still running 3k ppd lower than the 560Ti cards. Not quite as poorly as mk96 with his 10k off of it, but still underachieving.

Woah... wait a minute... Even with core version 2.22, there's no way a GTS-450 is getting anywhere near 20,000ppd on p8074, & core version 2.25 runs about 30-40% slower on Fermi cards. I think you have your numbers backwards. 26,149ppd is about where the 660Ti should be, 10,204ppd is about where the GTS-450 should be.

On Fermi cards, GPU ppd scales pretty evenly with the number of Cuda cores. At 92 cores, my GT-430 gets about 6,000ppd for about 65ppd per cuda core, At its best, my GTX-460 with 336 cores is getting 2min 56sec per frame for 19,017ppd (average is 17,600 - not sure why the big gap between best & average outputs...) for about 56.6ppd per cuda core. Lets assume that my 460 is performing poorly & my GT-430 is performing about where it should, given the clock speed difference between the 430 & 450, that should put a 450 between 68 & 76ppd per Cuda core depending on overclock. Going with the high-end, your 450 has 192 cuda cores time 76 = 14,592ppd. But even going nuts & assuming you're getting an astounding 100ppd per cuda core, that still leaves you just 19,200ppd...

im going on what the program is telling me.




and



they are running different projects right now though. but the ppd is still about the same as it was when they were running the same ones. Perhaps the program has mixed up which one is actually which.

it appears that the program is in fact showing the wrong cards in the client. I paused the one that claims to be the 660ti and MSI afterburner showed that the 660ti kept on going, and the 450 stopped.

mystery solved.

I had the same issue when I first went to the v7 client. I thought it was just a quirk of my system since no one else was reporting a similar situation.

bbies1973 just curios, you say the 660 is getting about 3k lower than the 560Ti, but what was the 560Ti getting & what is the 660 getting now?

That Zotac I got my hands on is a folding monster It's working on a 7623 work unit, some pep's are having issues with these work units & their overclocks. I've got the Zotac clocked as high as PrecisionX will allow & it's not even breaking a sweat. At a room temp of 76.7F, with overclock to 1215MHz core, running at 100% TDP, & 74% fan speed (also highest PrecisionX is allowing), it's only maxing out at 55C & averaging 6min 53sec per frame for 29,482ppd

That isn't compared specifically to the card fhat it replaced, but compared to two other 560Ti cards still running in my network. It also was the same difference across three different WUs.

so... your 560 Ti's are getting better than 30k ppd???

Since you've yet to post what your cards are getting on average, either frame rates or ppd, I can only speculate, but one more thing did come to mind...



if you're overclocking, make sure that your card is running steady at your overclock. Use PrecisionX, Afterburner, GPUz, or other monitoring software to check your clocks. The best overclock is one where your core stays at a constant speed like mine, note the core speed graph in PrecisionX. If you're overclocked too much and/or not cooling your GPU enough, the 600 series will change the core speed on the fly resulting in lower overall performance.

Mmm...I want a 660 now

Due to size/heat constraints I only got a 650 Ti to put in my server (also pulls double duty as my basement gaming PC), still getting 20K PPD out of it though, only got it up to 1019 core so far though (~10% over stock), hasn't gone over 58C when folding or gaming, anyone know what's safe temperature-wise for 24/7 for the GK106?

Officially 105C... Though I'd never be comfortable with more than 75C for 24/7 operation.

Thanks, that makes sense to me. Looks like I've got some experimenting to do, wonder how much more I can get out of it without messing with voltage (trying to be conservative with the new toy for now )