Folding Forum

Hi all, so I just got a gtx 770 to replace a couple of 460s, but it's not going so well. First off, I have a gt 520 to drive my monitor, and I don't fold on it, so the gtx 770 is unconnected to any video. So in FAHControl, (7.3.6) it shows the gt 520 and the gtx 770, but when I shut down the gt 520, my video got laggy, and when I checked the gpu usage in MSI Afterburner, it looked like the 770 usage was very low and the gt 520 usage was super high. So in FAHControl, I shut down the gtx 770 and turned on folding with the gt 520, and now things are going great, Afterburner says the 770 usage is high, things are great. But it looks like the gpu types in FAHControl are flipped. Has anyone else ever seen something like that?

Now in other news, it looks like the gtx 770 ppd is lower than I would have thought. The FAHControl estimated ppd is around 25k ppd, and I have a couple of 560 Ti cards in another machine that get roughly that. It's a 7660 wu, which I guess usually come in a little low, but still, seems weak. Afterburner is telling me that the gpu power is around 53%, and the temperature is 59 degrees, so there's plenty of overhead, apparently. So it looks like it's being underutilized, anyone know how to give it a kick up? Thanks!

It is relatively normal for v7 to mix up multiple video cards. The BIOS enumerates the cards, Then Windows enumerates, followed by the Windows drivers, then Cuda/OpenCL enumerates, and finally folding enumerates and by the time v7 gets to it the enumerations have little to do with what card is in what slot: it is ridiculous. If you ever update the Windows drivers or change cards the various components get an new opportunity to enumerate and with two cards you have a 50:50 chance the order will change.

That being said, you can manually configure the enumeration of each and every GPU slot by running Advanced Control followed by clicking configure and choosing the slots tab and double clicking the individual slot. Then manually assign the gpu-index for the slot. -1 means v7 auto-chooses and we already know that is wrong so just assign it to 0 (the first GPU) save and restart folding. Now test to see what card is actually folding, if it is the wrong card the repeat the above and choose 1 (the second card) instead. If you have multiple GPU slots you will have to manually configure all of them making sure that no two slots use the same gpu-index number or you'll end up with multiple WU's all running on the same video card.

P5-133XL wrote:... you can manually configure the enumeration of each and every GPU slot by running Advanced Control followed by clicking configure and choosing the slots tab and double clicking the individual slot....

... or you can just remember that they're switched and ignore the text string that identifies them.

cordis wrote:...Now in other news, it looks like the gtx 770 ppd is lower than I would have thought. The FAHControl estimated ppd is around 25k ppd, and I have a couple of 560 Ti cards in another machine that get roughly that. It's a 7660 wu, which I guess usually come in a little low, but still, seems weak. Afterburner is telling me that the gpu power is around 53%, and the temperature is 59 degrees, so there's plenty of overhead, apparently. So it looks like it's being underutilized, anyone know how to give it a kick up? Thanks!

Assuming that you are not having the client-type=advanced set, your GTX 770 (Kepler) is getting FahCore_15 WUs. Do note that FahCore_15 isn't optimized for Keplers. However, FahCore_17 is optimized for OpenCL GPUs and was recently released to the public using the client-type=advanced in the GPU Slot. I would suggest that you can try changing the client-type for your GTX 770 which should give you a nice PPD boost. To change the client-type, please use Advanced Control (AKA FAHControl) -> Configure -> Slots -> GPU (the correct one for GTX 770) -> Extra slot options -> Add:
Name: client-type
Value: advanced

I'm having a similar problem with a 8800 GT and GTX 460: viewtopic.php?f=80&t=24460

I can get the 460 folding fine but the minute I put the 8800 in, neither card folds.

In the above mentioned methods of adding slots, how should I set the opencl-index and cuda-index?
For the 460 it's set to gpu-index -1, opencl-index -1, cuda-index -1, not sure how I should set the 8800.

Thanks

I may have made things more complicated for myself, but there you go. So anyway, I figured if I was going to uninstall FAH, then I'd take the opprotunity to upgrade the machine, so now it's the single slot gt 520 providing the video, a gtx 560 ti I pulled from another machine, and the gtx 770. FAHBench gave me this list of cards:

GeForce GT 520
GeForce GTX 560 Ti
GeForce GTX 770

And that's for the Platform choices of OpenCL or CUDA (although the CUDA benchmarks didn't work). So I'm guessing that this means that the 520 is index 0, the 560 is index 1, and the 770 is index 2. So I added some slots, and they wound up looking like this:

<slot id='0' type='CPU'/>
<slot id='1' type='GPU'>
<cuda-index v='1'/>
<gpu-index v='2'/>
<opencl-index v='1'/>
</slot>
<slot id='2' type='GPU'>
<client-type v='advanced'/>
<cuda-index v='2'/>
<gpu-index v='0'/>
<opencl-index v='2'/>
</slot>

Where slot 1 should be the 560 ti, and that's how it's listed in fah control, and slot 2 should be the 770 but it's still listed as the 520. So anyway, folding with this, I started things up late last night, and they seemed to get going to a slow start. I checked again this morning, and it looks like both gpu slots were running on the 560 ti, the 770 doesn't seem to be getting used at all. I'm not sure how that's happening.
So I checked the documentation, and it looks like there are also gpu-vendor-id and gpu-device-id options. Should I try using those? I found the 770 in the GPUs.txt file that got downloaded, is that where the values come from? Or can I get them from gpu-z? At this point, I'm going to try to let the current WUs finish up, one at a time, and then delete and reconfigure the slots, and probably take out the cuda and opencl indicies, any other suggestions?

You should not need to change cuda-index or opencl-index: Those are used to enumerate Nvidia vs. AMD cards which is irrelevant because you have all Nvidia. It is best if you keep those at -1.

The only parameter you need to ID is gpu-index and if the auto setting did not properly determine which card goes with with slot then you are going to have to manually set it. The first will be 0 and then will increment from there. I can not predict what numbers each will need to be so you'll have to do trial & error to figure out which card is in each slot.

The gpu-vendor-id and gpu device-id are also things you shouldn't play with. All the cards will be identified by unique numbers in their bios and then compared to numbers contained in GPUs.txt, Only if your cards are not contained in the GPUs.txt would we need to go further to add them to the list. Explicitly adding the numbers to the client will not work because the servers need the information too and they won't get it from the client but rather their original GPUs.txt file stored on them.

You may help us if you showed the system + config portion of the log to see if all your cards were properly identified. You can copy & paste this portion of the log by going to the log tab and click refresh then schrool to the top.

IIRC, client v7.3.6 gives some "unrecognized option" warning for gpu-vendor-id and gpu-device-id. I suppose they aren't supported anymore.

Napoleon wrote:IIRC, client v7.3.6 gives some "unrecognized option" warning for gpu-vendor-id and gpu-device-id. I suppose they aren't supported anymore.

Right. If your GPUs.txt has been updated so that it matches what the server will accept, setting it locally doesn't allow you to override either the hardware detection or the server's white/black determination so they would serve no purpose if they did work Removing them makes sense, although I didn't remember that they did that.

Correct, the options for gpu-vendor-id and gpu-device-id are now automated and replaced by the GPUs.txt file which updates automatically.

However, the index settings (opencl, cuda, etc) are still valid and the way to correct this.

Hey, just an update, I got the slots set up as advanced without any cuda or opencl indicies, and although the 770 slot is still misidentified as a 520, it's now getting 8900 WUs with core 17 and generating ~85k ppd, so that's much better. Definitely nice to see the high ppd, I'll have to recalculate my ppds/watt and see how to readjust things. Might have to rethink my slower bigadv system. But first, I'll have to try getting gpus folding under linux.

One thing to remember is that Core_17. like Core_16, requires pretty nearly all of one (virtual) CPU, so if your machine is optimized for SMP, deallocate at least one CPU per GPU.

A related question, but not exactly. I installed a second GPU and wondered what the configuration setting should be for folding with two GPUs. Anyone know what it is?

Sailer wrote:A related question, but not exactly. I installed a second GPU and wondered what the configuration setting should be for folding with two GPUs. Anyone know what it is?

The first time I did a second GPU, I just uninstalled FAH, reinstalled it, and all worked great

It's actually pretty simple. You just add a new GPU slot through FAH control. If you let it default (-1), it will pick up the new GPU and all will be good.

If you are using the "client-type advanced" with your GPU and picking up Core_17, then you will want to reduce the number of cores that your CPU/SMP configuration uses so that you can get the most out of the GPUs in the machine. You can also do this through the slot configuration in FAH control.

-N0OA