I've noticed some 10629 WUs come through that seem to use a fluctuating percent of GPU time rather than being pegged at 99%.  Is this new to 10629 or have I simply not noticed this before?

Two 480 FTWs in SLI.  They are overclocked, but I'm not getting machine unstable errors and dropping the clocks doesn't seem to help.  The fluctuations are pretty dramatic as seen here.   Also, deleting the WU and starting a new one usually pegs the same GPU back to 99%.

Yessir, I'm on an E760 Classy... and only because I can't fit the 4-way board in my case ;-)   SLI is on at the moment.

My IOH vcore is @ 1.4v per the x58 OC guide in the mobo forum.  I'm more than willing to try bumping that around a bit, but whatever is happening seems to have just started with these WUs.  Also, I confirmed that deleting a WU and grabbing a new one kills the problem instantly...  so I wonder if some WUs have something slightly different about them that causes the fluctuation?  They are all 10629.

Next time I get a fluctuating WU I'll try a cold boot w/ the same WU to see if it persists across boots.

Thanks for the ideas/help.

It only happens on GPU2...  I thought it might be a problem with the GPU, but I took your advice and disabled SLI and the problem went away.  Just to be absolutely sure, I re-enabled SLI and the problem came back. 

Very weird, since I thought that the FAH GPU client addresses the GPUs individually so that SLI wasn't a part of the equation.  Also weird that it seems like it just started happening very recently.

Anyway, thanks -- at least I have a workaround, even if I don't totally understand what's going on!

Yeah, this problem (as you can see on the graph) was more drastic than that...  the usage would swing between 20% and 100% and it seemed like it was taking that much longer to finish those WUs.  I have yet to discover if the problem is associated with the latest project or if the problem is associated with the latest Nvidia betas, both of which landed for me at about the same time.  I'll just try to remember to disable SLI when I'm done with the computer for the night for now.

There does seem to be a clear relationship between certain WUs and utilization though (even if it only happens in SLI), since the problem persists between reboots and restarts of the GPU client and the only way to make it stop is to get a new WU.  Sometimes I have to abort/delete more than one new WU to get one that doesn't exhibit this behavior, but the problem is either 100% there or 100% not there for the life of the WU.

rambler358
What Nvidia drivers are you using?

Latest betas...  257.21.  Think I should try downgrading to the prior 25x driver to see if the problem persists across drivers?

chungenhung

i thought there were workunits that don't use 100% GPU. I saw some discussion about it on foldingforums.org

We did see a bunch up to a couple of months ago, but not lately, and definitly not on the new GPU3 setup.

I was having this issue all the time on the GPU driving my display. I figured out that if you raise the priority level of that core above the other one, i.e. from "low" to "below normal" in the task manager, they will be both at 99% all the time. I am running SLI enabled also.

itsmekirill

I was having this issue all the time on the GPU driving my display. I figured out that if you raise the priority level of that core above the other one, i.e. from "low" to "below normal" in the task manager, they will be both at 99% all the time. I am running SLI enabled also.

I just saw your reply.  I'll definitely give this a shot!  To be honest I haven't noticed much performance degradation in the games I'm playing at the moment w/ SLI off, but I'd really like to just leave it on if possible.  Set it and forget it!