Dual SLI GPU Failure Issues [solved]

[Devlin85]

I have 2 graphics cards, one I want to use for folding (EVGA 650 ti) and a card to be used for handling the desktop graphics and watching video (GT 610) while the other is dedicated to folding. The issue I'm having is when the 650 Ti starts folding it completely slows the system to a chug.. using the computer becomes literally impossible.. It's only using 25% cpu, nothing is hot (40-50c), I can't explain this or figure out what to do. My other system I have has 2 760 GTX's and even when both of those cards are folding I can use the PC without a skip, so this is really confusing me. I have tried switching slots (one runs in 2x, one is full) switching all of the CUDA settings to specify what programs go to what card in nvidia control.. Nothing is helping.

Mod edit: Topics merged. Title edited.

[Rel25917]

Have you checked with something like gpu-z or evga precision to make sure the right card is being used by FaH? Sounds like it's starting up on the wrong card and causing screen lag.

[P5-133XL]

Could you please post the folding log, including the system and config portions? It would be useful.

Are you using the most recent client v7.4.4? Is your monitor plugged into the GT 610? If so, using folding's Advanced Control identify the slot that is attached to the 610 and right click on it then select on idle. That should help a lot.

If your monitor is not plugged into the GT 610, then you may need to fiddle with the Nvidia Control Panel so that that the GT 610 is used as the primary display. Power down, switch the display over to the 610, and power back up. Finally, go to the Folding's advanced control panel and identify the slot used by the GT 610 and right-click and select on idle.

You can control the idle time by assigning a screen saver or adjusting the display power-down setting. That way you should not be experiencing much lag because the video card that runs the display is only folding when you are not using the machine.

If the folding client is confused as to which video card is the 610 and which is the 650 it is possible to manually switch them as far as the folding client is concerned but that will likely require some fiddling.

[Devlin85]

yeah.. wow.. I guess F@h is really stupid sometimes.. it has the names/slots switched.. told it to use the 610 and now it's actually using the 650 ti.. yeesh! thank you!

[Devlin85]

And that solved one problem and led to another.. it's pulling the wrong cores and it doesn't like that at all!

[Devlin85]

I tried with a GT 210 too.. same problem.. has the slots backwards.. now it's trying to work a core 11 WU on a 650 ti.. what a mess...

[bollix47]

Have a look at this post to see if it can help with your setup.

[P5-133XL]

You can fiddle with extra client options: gpu-index, opencl-index, and cuda-index on each slot to manually reorder the slots, if the client gets them wrong.

[Devlin85]

Thanks P5-133XL! Feel like an idiot for not trying that first, never had cards do that b4. It looks like the problem I was having was actually: no matter what slot I choose it completely ignored the 760 ti and used the gt 210 instead (gpu index 0 or 1) But luckily the cuda/opencl index force worked. The folding card is in my computer slot 0, but for some reason shows up as slot 1 in f@h (I can only assume it's an alphabetical thing as Nvidia control panel does it as well) and I had to make it gpu index slot 1, cuda 0, opencl 0. Looks like I'm golden now.

[P5-133XL]

The vast majority of times, the client gets it right. The biggest problem is when a card is added (or drivers change) after the client was configured. When either of those two occur, both the BIOS and Windows re-enumerate the cards and it can totally mess up the folding client. I have not seen any pattern to re-enumeration when it occurs. I have had a stable folding system, changed the drivers and had the order of my cards switched with no HW changes. Whatever it is, it is not alphabetical because with no HW changes the order would not have changed.

[Devlin85]

I stuck with the 327.23 driver after I concluded no drivers made a difference. I even took both cards out, added the new card by itself, and then re added the smaller card for system afterwards in the second slot, I don't know why but it always ended up with the GTX 760 listed on the bottom (which is alphabetical by card name in nvidia cpanel and matches the alphabetical position in device manager under display adapters) even when it was in the system first and in slot 0. So I really can't explain it. Just definitely setting everything manually from now on.

I added the card in my Media Center PC, 4 tuner HD Cable Card Reader, i5-2500k, 12GB DDR3, 2TB Hard Drive, and now folding and recording 24/7 with no video lag! 34k PPD right now. Not bad at all! Not like my 2 760's, but Best Buy put all their 650 Ti's on clearance for 104.99!! Couldn't pass that up.. no way!

Thanks for the help.

[bruce]

If all else fails, uninstall FAH completely and then install the latest version. Let me know if this sets everything up correctly.

Getting FAH to recognize the right GPU when there is more than one has been a problem for a long time. Microsoft has their own way of recognizing multiple GPUs and Linux/MacOS use another method. Over the development history of FAH, there have been several changes which have improved this situation but I'd like to know if it fixed your problem. As P5 mentioned, it's one thing to get it right on a fresh install and quite another if GPUs are changed after FAH is already running.

[Devlin85]

I have had a lot of issues processing the 9101 WU's and only in one card. I am using 2 EVGA GTX 760 Superclocked's and at first it was only those 9101's but then as of recent more and more started to occur. I ended up figuring out the issue was with the boost clock on one of the cards and as it was a card added to the system after the original install it was impossible to figure out which physical card it was due to the identical names and the way the system lays out the cards in windows. But after much observation (and it might not be the same for everybody) I figured out that one of my cards was jumping beyond the standard boost clock, my cards are supposed to run at 1149MHz boost clock, and at 1079MHz standard but for some reason 1 card was jumping up to 1188MHz at random and this was the problem the whole time! So a little Precision X mod (-39 Mhz) and no errors on any projects in days!! Numerous posts and this was never suggested.

A lesson to check you specs and make sure your cards are performing as intended.