2 GPUs being identified as each other

[Annihilator7]

I recently added a Nvidia 660 Ti to my computer, but the client is having problems telling it apart from my 275.
I have 3 GPUs: first is the 660 Ti, second is a 460, third is a 275.
SMP folding and the 460 both work fine.
I have added the latest GPUs.txt

The folding client is mis-identifying the 275 as my 660 Ti, and the 660 Ti as my 275:

When I run the folding slot that identifies itself as my 275, my 660 Ti's memory usage rises slightly, and its temperature increases, but without any usage %. No activity on the actual 275.

When I run the folding slot that claims to be my 660 Ti, the 275's memory usage increases, the 275's memory usage and GPU usage rise, along with the temperature of course. The GPU usage drops off after roughly 15 seconds, although the memory stays elevated.

Is there a way I can fix this?

[7im]

Hello Annihilator7, welcome to the forum.

Yes, there is. We can manually set the device order. What is the order of the GPUs as listed at the bottom of the GPU-Z application?

[Annihilator7]

660 Ti
460
275

[bollix47]

When you click on Configure and then the slots tab you should see 3 gpu slots. Click on the 660 Ti and change the gpu-index to 0 from -1. Click on ok and save.

Do the same for the 460 but use 1 for the gpu-index instead of -1.

And for the 275, same procedure but use 2 for the gpu-index.

Try the above and see if the identification problem is resolved.

Because of the problem with the client identifying with the wrong gpu the client might have downloaded the wrong type of work(for the 660 and the 275) and you might have to delete the appropriate work queue folders in the work directory.

e.g. If the 660 is using Work Queue 00 (not necessarily slot 00) then you would delete a folder called 00 under work in the Data Directory.

[Annihilator7]

Problem is persisting. The slot using GPU index 0 says it has the 275, and the slot with GPU index 2 thinks it has the 660 Ti. Index 1 works fine for the 460.

Could changing the GPU Core Indices be a solution here? The description of those in the program sounds like the kind of issue I'm having, but I haven't been able to find anything about them.

[bollix47]

AFAIK they should be in the same order as GPU-Z but I don't have a computer with multiple GPUs to test that.

Did you restart the client after making the changes?

If not, then please do so.

If you did restart and they still appear switched then reverse the gpu-index for those two and restart again.

[Annihilator7]

Tried swapping them back and forth, with client restarts, but problem persists. I don't think the GPU index alone will fix this, as after swapping it around, the slots just swap what card they think they have back to the wrong one. Is it possible, after telling slot 00 to use GPU index 0, to then force it to treat its card as a 660 Ti even if it's (wrongly) detecting that the card is the 275? (And vice versa for the other slot/card)

[bollix47]

Have a look at this thread:

viewtopic.php?f=67&t=22103&p=220716&hilit=multiple+gpus#p220716

In particular this post in that thread:

viewtopic.php?p=220714#p220714

It appears that just changing the gpu-index is not enough.

Also if you could refresh your log and copy/paste the System Info section plus the config section at the top of the log it may be helpful.

[Annihilator7]

So changing the GPU Core Indices is the right fix, then. I'll poke around with that and report the results.

Code: Select all

16:04:07:******************************* System ********************************
16:04:07:          CPU: Intel(R) Core(TM)2 Quad CPU Q9550 @ 2.83GHz
16:04:07:       CPU ID: GenuineIntel Family 6 Model 23 Stepping 10
16:04:07:         CPUs: 4
16:04:07:       Memory: 8.00GiB
16:04:07:  Free Memory: 5.85GiB
16:04:07:      Threads: WINDOWS_THREADS
16:04:07:   On Battery: false
16:04:07:   UTC offset: -4
16:04:07:          PID: 5112
16:04:07:          CWD: C:/Users/Carl/AppData/Roaming/FAHClient
16:04:07:           OS: Windows 7 Professional
16:04:07:      OS Arch: AMD64
16:04:07:         GPUs: 3
16:04:07:        GPU 0: NVIDIA:1 GT200b [GeForce GTX 275]
16:04:07:        GPU 1: FERMI:1 GF104 [GeForce GTX 460]
16:04:07:        GPU 2: NVIDIA:3 GK104 [GeForce GTX 660 Ti]
16:04:07:         CUDA: 3.0
16:04:07:  CUDA Driver: 5000
16:04:07:Win32 Service: false
16:04:07:***********************************************************************
16:04:07:<config>
16:04:07:  <!-- FahCore Control -->
16:04:07:  <checkpoint v='10'/>
16:04:07:  <cpu-usage v='70'/>
16:04:07:
16:04:07:  <!-- Folding Slot Configuration -->
16:04:07:  <gpu v='true'/>
16:04:07:
16:04:07:  <!-- Network -->
16:04:07:  <proxy v=':8080'/>
16:04:07:
16:04:07:  <!-- User Information -->
16:04:07:  <passkey v='********************************'/>
16:04:07:  <team v='111065'/>
16:04:07:  <user v='Annihilator7'/>
16:04:07:
16:04:07:  <!-- Folding Slots -->
16:04:07:  <slot id='3' type='SMP'/>
16:04:07:  <slot id='1' type='GPU'>
16:04:07:    <gpu-index v='2'/>
16:04:07:  </slot>
16:04:07:  <slot id='0' type='GPU'/>
16:04:07:  <slot id='2' type='GPU'/>
16:04:07:</config>

[Annihilator7]

Well, I've got mixed results. GPU index 0 will fold the 275 when given opencl-index and cuda-index of 2. I've tried GPU index 2, the 660 Ti, with Core Indices from 0 to 4, but none of them work. Index 1 is the 460.

Have others been folding the 660 Ti successfully?

[bollix47]

Can you try using 3 in opencl-index and cuda-index for the 660, then post the log showing it's attempt at folding.

[P5-133XL]

Currently, the 660 will require membership in the beta team to fold with. Kepler GPU's require a specialized core that is not yet avail to the general public.

[bollix47]

Also have a look at the following thread where another folder has a 660ti. You could check your ids using GPU-z and make sure the ids for your model are indeed in GPUs.txt.

viewtopic.php?p=222988#p222988

[Annihilator7]

I filtered this to only include Slot 1 info.

Code: Select all

*********************** Log Started 2012-09-02T17:38:35Z ***********************
17:38:36:WU01:FS01:Starting
17:38:36:WU01:FS01:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" C:/Users/Carl/AppData/Roaming/FAHClient/cores/www.stanford.edu/~pande/Win32/AMD64/NVIDIA/Fermi/Core_15.fah/FahCore_15.exe -dir 01 -suffix 01 -version 701 -lifeline 5676 -checkpoint 10 -cpu 70 -gpu 3
17:38:36:WU01:FS01:Started FahCore on PID 4404
17:38:36:WU01:FS01:Core PID:3256
17:38:36:WU01:FS01:FahCore 0x15 started
17:38:36:WU01:FS01:0x15:
17:38:36:WU01:FS01:0x15:*------------------------------*
17:38:36:WU01:FS01:0x15:Folding@Home GPU Core
17:38:36:WU01:FS01:0x15:Version                2.22 (Thu Dec 8 17:08:05 PST 2011)
17:38:36:WU01:FS01:0x15:Build host             SimbiosNvdWin7
17:38:36:WU01:FS01:0x15:Board Type             NVIDIA/CUDA
17:38:36:WU01:FS01:0x15:Core                   15
17:38:36:WU01:FS01:0x15:GPU device info vendor=0 device=0 name=NA match=0 deviceId=3
17:38:36:WU01:FS01:0x15:
17:38:36:WU01:FS01:0x15:Window's signal control handler registered.
17:38:36:WU01:FS01:0x15:Preparing to commence simulation
17:38:36:WU01:FS01:0x15:- Looking at optimizations...
17:38:36:WU01:FS01:0x15:- Files status OK
17:38:36:WU01:FS01:0x15:sizeof(CORE_PACKET_HDR) = 512 file=<>
17:38:37:WU01:FS01:0x15:- Expanded 66228 -> 285182 (decompressed 430.6 percent)
17:38:37:WU01:FS01:0x15:Called DecompressByteArray: compressed_data_size=66228 data_size=285182, decompressed_data_size=285182 diff=0
17:38:37:WU01:FS01:0x15:- Digital signature verified
17:38:37:WU01:FS01:0x15:
17:38:37:WU01:FS01:0x15:Project: 8010 (Run 0, Clone 1133, Gen 87)
17:38:37:WU01:FS01:0x15:
17:38:37:WU01:FS01:0x15:Assembly optimizations on if available.
17:38:37:WU01:FS01:0x15:Entering M.D.
17:38:38:WU01:FS01:0x15:Tpr hash 01/wudata_01.tpr:  1464577836 4212705263 383786006 2339184692 1846143707
17:38:38:WU01:FS01:0x15:GPU device info: vendor=0 device=0 name=<NA> match=0
17:38:38:WU01:FS01:0x15:Working on Good ROcking Metal Altar for Chronical Sinners
17:38:38:WU01:FS01:0x15:Client config unavailable.
17:38:39:WU01:FS01:0x15:Starting GUI Server


Well that's interesting...

Code: Select all

17:38:38:WU01:FS01:0x15:Working on Good ROcking Metal Altar for Chronical Sinners

Sounds like science to me.

Currently, the 660 will require membership in the beta team to fold with. Kepler GPU's require a specialized core that is not yet avail to the general public.

Problem found, then.
I'll assume there's no ETA/guesstimate on public availability of this. Should I look into joining that beta?

[P5-133XL]

That would be a reasonable solution to that specific problem. Joining the beta team is open to all if you are willing to agree to their conditions. Beta Team - Rules Of Engagement

Stanford has a policy of never giving ETA's, so I can't help you there.