Folding ONLY uses a specific GPU, regardless of setting.

[Azathoth321]

I have tried setting up the F@H client to run solely on a dedicated GTX 750Ti.
My PC has Two GTX 970's as well, but I do not wish to fold on these.

The issue is that the client will ALWAYS run on my second GTX 970 (Index 2) no matter what setting I try, as long as it's a single workunit. (If I have 3 GPU workunits, all 3 GPU's are used.)
If I manually set the Index to 0 (GTX 750Ti) In the console it says it is starting a unit on the GTX 750Ti, but monitoring programs show my second GTX 970 being utilized still.

I even attempted removing the GTX 970 from the GPU.txt in an attempt to...do something. ---Nothing

Any ideas?

[Gooders]

Have you got a copy of your log you can post?

[Azathoth321]

Code: Select all

*********************** Log Started 2014-11-09T02:00:06Z ***********************
02:00:06:************************* Folding@home Client *************************
02:00:06:      Website: http://folding.stanford.edu/
02:00:06:    Copyright: (c) 2009-2014 Stanford University
02:00:06:       Author: Joseph Coffland <joseph@cauldrondevelopment.com>
02:00:06:         Args: 
02:00:06:       Config: C:/Program Files (x86)/FAHClient/config.xml
02:00:06:******************************** Build ********************************
02:00:06:      Version: 7.4.4
02:00:06:         Date: Mar 4 2014
02:00:06:         Time: 20:26:54
02:00:06:      SVN Rev: 4130
02:00:06:       Branch: fah/trunk/client
02:00:06:     Compiler: Intel(R) C++ MSVC 1500 mode 1200
02:00:06:      Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE
02:00:06:               /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl
02:00:06:     Platform: win32 XP
02:00:06:         Bits: 32
02:00:06:         Mode: Release
02:00:06:******************************* System ********************************
02:00:06:          CPU: Intel(R) Core(TM) i5-4670K CPU @ 3.40GHz
02:00:06:       CPU ID: GenuineIntel Family 6 Model 60 Stepping 3
02:00:06:         CPUs: 4
02:00:06:       Memory: 15.83GiB
02:00:06:  Free Memory: 11.69GiB
02:00:06:      Threads: WINDOWS_THREADS
02:00:06:   OS Version: 6.1
02:00:06:  Has Battery: false
02:00:06:   On Battery: false
02:00:06:   UTC Offset: -5
02:00:06:          PID: 5476
02:00:06:          CWD: C:/Program Files (x86)/FAHClient
02:00:06:           OS: Windows 7 Home Premium
02:00:06:      OS Arch: AMD64
02:00:06:         GPUs: 1
02:00:06:        GPU 0: NVIDIA:4 GM107 [GeForce GTX 750 Ti]
02:00:06:         CUDA: 5.2
02:00:06:  CUDA Driver: 6050
02:00:06:Win32 Service: false
02:00:06:***********************************************************************
02:00:06:<config>
02:00:06:  <!-- Network -->
02:00:06:  <proxy v=':8080'/>
02:00:06:
02:00:06:  <!-- User Information -->
02:00:06:  <user v='Azathoth321'/>
02:00:06:
02:00:06:  <!-- Folding Slots -->
02:00:06:  <slot id='0' type='GPU'/>
02:00:06:</config>
02:00:06:Trying to access database...
02:00:06:Successfully acquired database lock
02:00:06:Enabled folding slot 00: READY gpu:0:GM107 [GeForce GTX 750 Ti]
02:00:06:WU00:FS00:Starting
02:00:06:WU00:FS00:Running FahCore: "C:\Program Files (x86)\FAHClient/FAHCoreWrapper.exe" "C:/Program Files (x86)/FAHClient/cores/web.stanford.edu/~pande/Win32/AMD64/NVIDIA/Fermi/Core_17.fah/FahCore_17.exe" -dir 00 -suffix 01 -version 704 -lifeline 5476 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
02:00:06:WU00:FS00:Started FahCore on PID 8024
02:00:06:WU00:FS00:Core PID:8068
02:00:06:WU00:FS00:FahCore 0x17 started
02:00:06:WU00:FS00:0x17:*********************** Log Started 2014-11-09T02:00:06Z ***********************
02:00:06:WU00:FS00:0x17:Project: 9201 (Run 270, Clone 4, Gen 8)
02:00:06:WU00:FS00:0x17:Unit: 0x0000000f6652edc45399e0a5ae929d32
02:00:06:WU00:FS00:0x17:CPU: 0x00000000000000000000000000000000
02:00:06:WU00:FS00:0x17:Machine: 0
02:00:06:WU00:FS00:0x17:Digital signatures verified
02:00:06:WU00:FS00:0x17:Folding@home GPU core17
02:00:06:WU00:FS00:0x17:Version 0.0.52

[davidcoton]

That portion of the log shows that FAH is only recognising one GPU -- the 750Ti. It only shows one WU, again on the 750Ti. Do you have a log showing multiple WUs running? Please post it using code tags in the full editor.

[Napoleon]

Editing GPUs.txt is not a good idea / won't help. Try setting cuda-index/opencl-index manually for the GPU slot: viewtopic.php?p=199379#p199379

[7im]

Editing GPUs.txt is not a good idea / won't help. Try setting cuda-index/opencl-index manually for the GPU slot: viewtopic.php?p=199379#p199379

If you have added or remove a GPU since installing the FAH client, you need the uninstall and reinstall it again as a first step. Then follow the guide to set the indexes.

[Azathoth321]

That portion of the log shows that FAH is only recognising one GPU -- the 750Ti. It only shows one WU, again on the 750Ti. Do you have a log showing multiple WUs running? Please post it using code tags in the full editor.

This was because I disabled it's ability to even recognize my 970's, because regardless of the console showing what it displayed as the working GPU, it would ONLY use the second 970.

Editing GPUs.txt is not a good idea / won't help. Try setting cuda-index/opencl-index manually for the GPU slot: viewtopic.php?p=199379#p199379

I tried doing this as well with no dice, however it reminded me there was a setting in the Nvidia control panel to untick checkboxes of which GPU's should be able to run CUDA applications. I simply unticked my GTX 970's and left only my 750Ti, and now it is correctly running on my 750Ti.

[davidcoton]

Editing GPUs.txt is not a good idea / won't help. Try setting cuda-index/opencl-index manually for the GPU slot: viewtopic.php?p=199379#p199379

I tried doing this as well with no dice, however it reminded me there was a setting in the Nvidia control panel to untick checkboxes of which GPU's should be able to run CUDA applications. I simply unticked my GTX 970's and left only my 750Ti, and now it is correctly running on my 750Ti.

That is bizarre. Core17 supposedly doesn't use CUDA, it runs OpenCL to provide cross-platform compatibility.

[7im]

I'm glad you found another workaround. If not, setting the indexes correctly is a known good solution as well.

[Azathoth321]

I'm glad you found another workaround. If not, setting the indexes correctly is a known good solution as well.

I did try this in my troubleshooting, setting the index made the console show it was using the correct GPU, but the PPD estimate and the GPU usage/Temperature was always pointing towards the second GTX 970 being used, regardless of console output. Strange.

Well if anyone else has a similar problem, be sure to check or uncheck which Graphics Processors can be allowed to use CUDA applications.

[heikosch]

I'm glad you found another workaround. If not, setting the indexes correctly is a known good solution as well.

I did try this in my troubleshooting, setting the index made the console show it was using the correct GPU, but the PPD estimate and the GPU usage/Temperature was always pointing towards the second GTX 970 being used, regardless of console output. Strange.

Well if anyone else has a similar problem, be sure to check or uncheck which Graphics Processors can be allowed to use CUDA applications.

F@H has some problems to detect the GPUs in multiple GPU systems, this is well known and documented problem. Don´t trust the F@H output!
Use a third party program like GPU-Z, MSI Afterburner, EVGA Precision etc. instead.

Heiko

PS: F@H core 0x17 uses OpenCL - not CUDA. (EDIT: added core Version, thanks to bruce)

[bruce]

PS: F@H uses OpenCL - not CUDA.

That is currently true, but not a reliable statement. Some FahCores, including Core_17 use OpenCL, but previous cores have used CUDA (e.g.- FahCore_15) and future cores might.

Apparently when you configure the drivers to disable CUDA, that also disables OpenCL, but if that's true, it's NVidia's way of doing things, not Stanford's.

[Napoleon]

Well if anyone else has a similar problem, be sure to check or uncheck which Graphics Processors can be allowed to use CUDA applications.

I'm under the impression that NV OpenCL is actually implemented on top of CUDA. If so, when excluding certain GPU(s) from CUDA in the driver, you'll exclude them from GPU computing altogether. In any case, a nice workaround.