GPU WU time issue - GTX 275 + GTX 465

Not sure what just happened here, but I was getting WU times of about 14 min + with the GTX 465 O/Ced to 800 and WU times of roughly 30min with the GTX 275 O/ced to 750-775.

Now, for some reason the GTX 275 WUs have literally doubled - they are at 1 hour. The GTX 465 WUs looked to have gone up a little as well, but I rebooted, and now the 465 WUs seem roughly back in line, maybe just a tad longer. Even after reboot, the 275 WUs are still at 1 hr to complete.

Any idea what might have occurred here?

Also, as an aside, does anyone have a good explanation why the 275 - pulling 219W and 1000GFLOPS stock - has such a poor showing against the 465 - pulling 200W and "only" 855GFLOPS stock? I would have thought the 275 would be a much stronger pulling card. Not sure what the cause of the huge disparity is, unless it has something to do with a particular CUDA and PG coding nuance, and not just "Raw GFLOPS" power.

What driver are you using? My immediate thought is that this is a driver issue. I think the newer Nvidia drivers are geared towards Fermi cards...

hmm. i thought core speed wasnt relevent? i heard from someone here to only oc the shader and you can even lower core speed and mem to lessen heat. on my 275 im at 648 core 1577 shader 1188 mem and i do w/u's in 32-34 min each. ive had it higher and can do them as fast as 31 minutes per so far. temps are in the mid to high 80's i took off my side panel and have a 8 in. fan blowing on card that gets it down to 70 degrees. dont know about your issue drpop but i do believe i have the latest drivers i allways am checking. good luck on that though.

http://www.primegrid.com/forum_thread.php?id=3339

found this maybe your card is throttleing? i oc with evga precision and oc and watch for the temps to drop then i know it didnt hold.

I forgot about that - good point, Greg. The new drivers throttle by default, so that might be an issue...

Interesting, thanks for the comments. I did try going back a version in the drivers, but didn't change anything. My temps are in the 70s or lower, o/c is with MSI Afterburner, and I can see both GPUs@ 99%. What is strange to me is that 1. Similar GFLOPS cards, but the 465 is twice as fast- so there must be something other than GFLOPS that matters for PG code. And, 2. I can't figure out why it was 30min per WU on the 275 and then just doubled to 1hr per! That is only 25% of what the 465 is doing, and that's not right[..]:-??. I wonder if they don't play well together...I could try seperating them, maybe?

Just send me the 275! :p;)

Thought you were out of room? Do you have an open pci-e now? ;)

No but I could sleep with it under my pillow. :cool:

@DP: Have you looked at the reports for the wu's? There might be a clue in them somewhere. Also check the batch #'s. Maybe project admins changed something?

Did you futz with the throttling setting that Greg linked to? That's just changing the power management setting within NCP, under 3D settings.

Another thing they mention is that it might be worth dedicating a core to the GPU...

Just a couple more thoughts...

PS: you did DriverSweep in between driver installations, yes?

Sorry for the late reply here - been busy. I will try it in Kat's rig and see what happens with just 1 CUDA card in there.

This may be the problem.


http://forums.nvidia.com/index.php?s...&#entry1237460

Thanks, that was an interesting read. :)

That post basically says.....Run the cards in SLI, don't spread out the desktop and all is well. That's with the 270 driver.

I'm running 266 driver, non-SLI, no expanded desktop and haven't had any issues.
OCing works, but can't seem to get it stay put as the clocks keep reverting to stock speed.
Other than that, I'm golden.

Is the 465 a card you flashed the BIOS on?

Yo-