#597 GPU Detection error: 2x Fermi should be 1xFermi + 1xg80

Moderators: Site Moderators, PandeGroup

#597 GPU Detection error: 2x Fermi should be 1xFermi + 1xg80

Postby LiLChris » Mon Mar 28, 2011 11:14 pm

Hopefully I am posting this in the right place. :oops:

I install the client in GPU mode it auto detects and starts folding, but my GPUs are mixed up.
Tried setting it up manually in expert/configure but its still wrong.

My setup
i7 930 - Evga e758 3 way - 6gb dominator
1st slot - GTX 480
2nd slot - 9800gtx+

Below is the log and you can see its backwards, and it seems to be wanting to fold both cards as a Fermi card.

Code: Select all
*********************** Log Started 28/Mar/2011-22:48:47 ***********************
22:48:47:************************* Folding@home Client *************************
22:48:47:      Website: http://folding.stanford.edu/
22:48:47:    Copyright: (c) 2009,2010 Stanford University
22:48:47:       Author: Joseph Coffland <joseph@cauldrondevelopment.com>
22:48:47:         Args: --lifeline 2784
22:48:47:       Config: C:/Program Files (x86)/FAHClient/data/config.xml
22:48:47:******************************** Build ********************************
22:48:47:      Version: 7.1.21
22:48:47:         Date: Mar 23 2011
22:48:47:         Time: 16:13:46
22:48:47:      SVN Rev: 2883
22:48:47:       Branch: fah/trunk/client
22:48:47:     Compiler: Intel(R) C++ MSVC 1500 mode 1110
22:48:47:      Options: /TP /nologo /EHa /wd4297 /wd4103 /wd1786 /Ox -arch:SSE2
22:48:47:               /QaxSSE3,SSSE3,SSE4.1,SSE4.2 /Qrestrict /MT
22:48:47:     Platform: win32 Vista
22:48:47:         Bits: 32
22:48:47:         Mode: Release
22:48:47:******************************* System ********************************
22:48:47:           OS: Microsoft Windows 7 Home Premium
22:48:47:          CPU: Intel(R) Core(TM) i7 CPU 930 @ 2.80GHz
22:48:47:       CPU ID: GenuineIntel Family 6 Model 26 Stepping 5
22:48:47:         CPUs: 8
22:48:47:       Memory: 5.99GiB
22:48:47:  Free Memory: 3.70GiB
22:48:47:      Threads: WINDOWS_THREADS
22:48:47:         GPUs: 2
22:48:47:        GPU 0: FERMI:1 G92 [GeForce 9800 GTX+]
22:48:47:        GPU 1: FERMI:1 GF100 [GeForce GTX 480]
22:48:47:         CUDA: 2.0
22:48:47:  CUDA Driver: 3020
22:48:47:   On Battery: false
22:48:47:   UTC offset: -4
22:48:47:          PID: 3136
22:48:47:          CWD: C:/Program Files (x86)/FAHClient/data
22:48:47:Win32 Service: false
22:48:47:***********************************************************************



Here is an image of what happens...
My GTX 480 is folding at 100% with the correct core, but the 2nd GPU is not.

PS. Couldn't get the flags to work I found on the new wiki.

Image
Image
LiLChris
 
Posts: 45
Joined: Wed Aug 11, 2010 5:29 pm

Re: GPU Detection

Postby bruce » Mon Mar 28, 2011 11:20 pm

The GTX480 is correctly configure as FERMI. The 9800GTX should not be Fermi.

Right-click on Folding slot 0 and select Pause until we can figure this out.

Unlike V6, flags are rarely useful in V7, and (as you've discovered) difficult to use. With very rare exceptions, everything goes into the configuration using FAHConrol to change things when necessary.

I've opened Ticket #597 for this problem.
bruce
 
Posts: 22331
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: GPU Detection

Postby LiLChris » Mon Mar 28, 2011 11:21 pm

Actually if I stop folding slot 0, it stops the Fermi.
So I paused slot 1 and the Fermi is still folding.


Image
Last edited by LiLChris on Mon Mar 28, 2011 11:25 pm, edited 1 time in total.
LiLChris
 
Posts: 45
Joined: Wed Aug 11, 2010 5:29 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby heydabop » Tue Mar 29, 2011 11:11 pm

I'm experiencing the same exact behavior with an NVIDIA GTX 470 and a GTS 250. It appears almost as if the GPU ID's are switched?
heydabop
 
Posts: 8
Joined: Tue Mar 29, 2011 11:10 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby bruce » Wed Mar 30, 2011 4:18 am

heydabop wrote:I'm experiencing the same exact behavior with an NVIDIA GTX 470 and a GTS 250. It appears almost as if the GPU ID's are switched?
OK, but is there really a bug here? Do all of the GPUs that could be folding get a slot created and do they work? The sequence in which they're detected really shouldn't matter ... or GPU 1 could be assigned to Slot 0 and GPU 0 could be assigned to Slot 1 ... as long as they fold.

Please post the hardware configuration section at the beginning of the log file. The FAHClient should detect two GPUs and if you used the GPU configuration procedure, there should be two GPU slots. If they're both are processing a WU, what projects are they.

In the OP, if I understand correctly, work is NOT being assigned to the GTX480 and it's continually retrying. It that happening to you too, heydabop?

Both of you: Please post a segment of the log showing repeated attempts to download a core or the successful startup of a WU. Post it both for slot 0 and for slot 1.
bruce
 
Posts: 22331
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby heydabop » Wed Mar 30, 2011 4:59 am

Work is being assigned to my GTX 470, however it's identified as the GTS 250 in FAHControl. Work isn't successfully being assigned to my GTS 250.
Code: Select all
23:01:59:Added folding slot
23:01:59:Added folding slot
23:01:59:Saving configuration to config.xml
23:01:59:<config>
23:01:59:  <!-- FahCore Control -->
23:01:59:  <checkpoint v='10'/>
23:01:59:  <core-priority v='low'/>
23:01:59:
23:01:59:  <!-- Folding Slot Configuration -->
23:01:59:  <gpu v='true'/>
23:01:59:
23:01:59:  <!-- Logging -->
23:01:59:  <verbosity v='3'/>
23:01:59:
23:01:59:  <!-- Remote Command Server -->
23:01:59:  <password v=''/>
23:01:59:
23:01:59:  <!-- User Information -->
23:01:59:  <passkey v='********************************'/>
23:01:59:  <team v='111065'/>
23:01:59:  <user v='heydabop'/>
23:01:59:
23:01:59:  <!-- Folding Slots -->
23:01:59:  <slot id='2' type='SMP'>
23:01:59:    <client-type v='advanced'/>
23:01:59:    <max-packet-size v='big'/>
23:01:59:  </slot>
23:01:59:  <slot id='0' type='GPU'/>
23:01:59:  <slot id='1' type='GPU'>
23:01:59:    <gpu-id v='1'/>
23:01:59:  </slot>
23:01:59:</config>


Here's what happens when the GTS 250 (which appears to be identified as the GTX 470) is assigned a work unit.
Code: Select all
04:49:10:Slot 01 unpaused
04:49:10:Starting Unit 01
04:49:10:Running core: C:/Users/Ross/AppData/Roaming/FAHClient/cores/www.stanford.edu/~pande/Win32/x86/NVIDIA/Fermi/Core_15.fah/FahCore_15.exe -dir 01 -suffix 01 -lifeline 5488 -version 701 -checkpoint 10 -gpu 1
04:49:10:Started core on PID 2524
04:49:10:FahCore 0x15 started
04:49:11:Unit 01:
04:49:11:Unit 01:*------------------------------*
04:49:11:Unit 01:Folding@Home GPU Core
04:49:11:Unit 01:Version 2.15 (Tue Nov 16 09:05:18 PST 2010)
04:49:11:Unit 01:
04:49:11:Unit 01:Build host: SimbiosNvdWin7
04:49:11:Unit 01:Board Type: NVIDIA/CUDA
04:49:11:Unit 01:Core      : x=15
04:49:11:Unit 01: Window's signal control handler registered.
04:49:11:Unit 01:Preparing to commence simulation
04:49:11:Unit 01:- Ensuring status. Please wait.
04:49:16:Unit 00:Completed  17000000 out of 50000000 steps (34%).
04:49:20:Unit 01:- Looking at optimizations...
04:49:20:Unit 01:- Working with standard loops on this execution.
04:49:20:Unit 01:- Previous termination of core was improper.
04:49:20:Unit 01:- Going to use standard loops.
04:49:20:Unit 01:- Files status OK
04:49:20:Unit 01:sizeof(CORE_PACKET_HDR) = 512 file=<>
04:49:20:Unit 01:- Expanded 43753 -> 171827 (decompressed 392.7 percent)
04:49:20:Unit 01:Called DecompressByteArray: compressed_data_size=43753 data_size=171827, decompressed_data_size=171827 diff=0
04:49:20:Unit 01:- Digital signature verified
04:49:20:Unit 01:
04:49:20:Unit 01:Project: 6801 (Run 3428, Clone 4, Gen 11)
04:49:20:Unit 01:
04:49:20:Unit 01:Entering M.D.
04:49:22:Unit 01:Tpr hash 01/wudata_01.tpr:  1256493998 2577052743 164322495 3906964246 1978304629
04:49:22:FahCore running Unit 01 returned: UNKNOWN_ENUM (-1)
04:49:22:WARNING: Unit 01 Too many errors, failing


Here's a succesful download and start of a GPU WU on the GTX 470 (marked as GTS 250 in FAHConfig)
Code: Select all
04:08:11:Starting Unit 00
04:08:11:Connecting to 171.64.65.64:8080
04:08:11:Running core: C:/Users/Ross/AppData/Roaming/FAHClient/cores/www.stanford.edu/~pande/Win32/x86/NVIDIA/Fermi/Core_15.fah/FahCore_15.exe -dir 00 -suffix 01 -lifeline 5488 -version 701 -checkpoint 10 -gpu 0
04:08:11:Started core on PID 7020
04:08:11:FahCore 0x15 started
04:08:12:Unit 00:
04:08:12:Unit 00:*------------------------------*
04:08:12:Unit 00:Folding@Home GPU Core
04:08:12:Unit 00:Version 2.15 (Tue Nov 16 09:05:18 PST 2010)
04:08:12:Unit 00:
04:08:12:Unit 00:Build host: SimbiosNvdWin7
04:08:12:Unit 00:Board Type: NVIDIA/CUDA
04:08:12:Unit 00:Core      : x=15
04:08:12:Unit 00: Window's signal control handler registered.
04:08:12:Unit 00:Preparing to commence simulation
04:08:12:Unit 00:- Looking at optimizations...
04:08:12:Unit 00:DeleteFrameFiles: successfully deleted file=00/wudata_01.ckp
04:08:12:Unit 00:- Created dyn
04:08:12:Unit 00:- Files status OK
04:08:12:Unit 00:sizeof(CORE_PACKET_HDR) = 512 file=<>
04:08:12:Unit 00:- Expanded 43443 -> 171827 (decompressed 395.5 percent)
04:08:12:Unit 00:Called DecompressByteArray: compressed_data_size=43443 data_size=171827, decompressed_data_size=171827 diff=0
04:08:12:Unit 00:- Digital signature verified
04:08:12:Unit 00:
04:08:12:Unit 00:Project: 6801 (Run 4715, Clone 4, Gen 1)
04:08:12:Unit 00:
04:08:12:Unit 00:Assembly optimizations on if available.
04:08:12:Unit 00:Entering M.D.
04:08:13:Unit 00:Tpr hash 00/wudata_01.tpr:  3010767663 2995436835 1886240533 2956735321 953810916
04:08:13:Unit 00:Working on ALZHEIMER'S DISEASE AMYLOID
04:08:13:Unit 00:Client config unavailable.
04:08:13:Unit 00:Setting checkpoint frequency: 500000
04:08:13:Unit 00:Setting checkpoint frequency: 500000
04:08:14:Unit 00:Starting GUI Server
04:08:17:Unit 03: 8.65%
04:08:23:Unit 03: 18.12%
04:08:29:Unit 03: 27.58%
04:08:35:Unit 03: 37.05%
04:08:41:Unit 03: 46.52%
04:08:47:Unit 03: 55.98%
04:08:53:Unit 03: 65.29%
04:08:59:Unit 03: 73.61%
04:09:05:Unit 03: 82.10%
04:09:11:Unit 03: 85.36%
04:09:17:Unit 03: 91.24%
04:09:23:Unit 03: 99.56%
04:09:24:Unit 03: Upload complete
04:09:24:Server responded WORK_ACK (400)
04:09:24:Final credit estimate, 1348.00 points
04:09:24:Cleaning up Unit 03
04:09:26:Unit 00:Completed    500000 out of 50000000 steps (1%).


Here's a screenshot of FAHConfig. Note that the selected WU is actually folding on the GTX 470, despite appearing to be identified as the GTS 250 in FAHConfig.
Image

I hope this helps.
heydabop
 
Posts: 8
Joined: Tue Mar 29, 2011 11:10 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby bruce » Wed Mar 30, 2011 5:45 am

That's only half of what I need. At the beginning of the log, there's a ***System*** section. See the part of the first post in this topic where LiLChris posted this segment of his log:
Code: Select all
22:48:47:******************************* System ********************************
22:48:47:           OS: Microsoft Windows 7 Home Premium
22:48:47:          CPU: Intel(R) Core(TM) i7 CPU 930 @ 2.80GHz
22:48:47:       CPU ID: GenuineIntel Family 6 Model 26 Stepping 5
22:48:47:         CPUs: 8
22:48:47:       Memory: 5.99GiB
22:48:47:  Free Memory: 3.70GiB
22:48:47:      Threads: WINDOWS_THREADS
22:48:47:         GPUs: 2
22:48:47:        GPU 0: FERMI:1 G92 [GeForce 9800 GTX+]
22:48:47:        GPU 1: FERMI:1 GF100 [GeForce GTX 480]
22:48:47:         CUDA: 2.0
22:48:47:  CUDA Driver: 3020
22:48:47:   On Battery: false
22:48:47:   UTC offset: -4
22:48:47:          PID: 3136
22:48:47:          CWD: C:/Program Files (x86)/FAHClient/data
22:48:47:Win32 Service: false
bruce
 
Posts: 22331
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby heydabop » Wed Mar 30, 2011 6:50 am

Oops, sorry. This?
Code: Select all
06:47:51:******************************* System ********************************
06:47:51:           OS: Microsoft Windows 7 Ultimate
06:47:51:          CPU: AMD Phenom(tm) II X6 1090T Processor
06:47:51:       CPU ID: AuthenticAMD Family 16 Model 10 Stepping 0
06:47:51:         CPUs: 6
06:47:51:       Memory: 8.00GiB
06:47:51:  Free Memory: 5.70GiB
06:47:51:      Threads: WINDOWS_THREADS
06:47:51:         GPUs: 2
06:47:51:        GPU 0: FERMI:1 G92 [GeForce GTS 250]
06:47:51:        GPU 1: FERMI:1 GF100 [GeForce GTX 470]
06:47:51:         CUDA: 2.0
06:47:51:  CUDA Driver: 3020
06:47:51:   On Battery: false
06:47:51:   UTC offset: -5
06:47:51:          PID: 3188
06:47:51:          CWD: C:/Users/Ross/AppData/Roaming/FAHClient
06:47:51:Win32 Service: false
heydabop
 
Posts: 8
Joined: Tue Mar 29, 2011 11:10 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby LiLChris » Wed Mar 30, 2011 9:00 pm

Yea heydabop, you have it backwards as well.
LiLChris
 
Posts: 45
Joined: Wed Aug 11, 2010 5:29 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby bruce » Wed Mar 30, 2011 11:39 pm

LiLChris wrote:Yea heydabop, you have it backwards as well.


I still don't understand what you're calling backwards.

22:48:47: GPU 0: FERMI:1 G92 [GeForce 9800 GTX+
22:48:47: GPU 1: FERMI:1 GF100 [GeForce GTX 480]
should say
22:48:47: GPU 0: NVIDA:1 G92 [GeForce 9800 GTX+]
22:48:47: GPU 1: FERMI:1 GF100 [GeForce GTX 480]
bruce
 
Posts: 22331
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby LiLChris » Thu Mar 31, 2011 1:16 am

Cause his GPU 0 should be the 470 and my GPU 0 should be the 480.
LiLChris
 
Posts: 45
Joined: Wed Aug 11, 2010 5:29 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby heydabop » Thu Mar 31, 2011 4:44 am

LiLChris wrote:Cause his GPU 0 should be the 470 and my GPU 0 should be the 480.

This.
My GTX 470 is in my first PCI-E slot and my GTS 250 is in my second PCI-E slot.
With the old F@H client (V6 I think), my GTX 470 was gpu 0 and my GTS 250 was gpu 1. Also, when running a WU on GPU 0 in this client (identified as the GTS 250), it runs on the GTX 470, not the GTS 250.
heydabop
 
Posts: 8
Joined: Tue Mar 29, 2011 11:10 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby bruce » Thu Mar 31, 2011 5:06 am

No, you're missing the point. It makes absolutely no difference which GPU is called GPU0 and which is called GPU1 (as long as the client is internally consistent about it). What does matter is that the GF100 is properly identified as FERMI:1 in the hardware detection report and that the G92 is properly identified as NVIDIA:1. The latter is not happening. See the part I marked in red three posts above.

What's really happening is that the hardware detection routine is misidentifiying a non-fermi GPU as a fermi GPU and then assigning Fermi work to it ... which fails. The NVIDIA:1 is essentially the same thing you get when you use the V6 flag -forcegpu nvidia_g80 but instead the V7 client is assuming the equivalent of a -forcegpu nvidia_fermi and applying it to both gpus, not just the one that is really a fermi.

They're working on the bug. We'll have to wait until they give us a new version of V7 that fixes the bug. That's why we beta test.

It does work correctly if you have a single GPU
viewtopic.php?p=181095#p181095
Adam A. Wanderer wrote:00:49:19: GPUs: 1
00:49:19: GPU 0: NVIDIA:1 G92 [GeForce GTS 250]
00:49:19: CUDA: 1.1
00:49:19: CUDA Driver: 3020
bruce
 
Posts: 22331
Joined: Thu Nov 29, 2007 10:13 pm
Location: So. Cal.

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby heydabop » Thu Mar 31, 2011 6:04 pm

I might be waay off here, but I think it does matter, because right now regardless of how they're identified, gpu 0 points to my GTX 470 and gpu 1 points to my GTS 250. So if the log says
Code: Select all
22:48:47: GPU 0: NVIDA:1 G92 [GeForce 9800 GTX+]
22:48:47: GPU 1: FERMI:1 GF100 [GeForce GTX 480]

All that's happening is that non-fermi WU's will be put on my GTX 470, while fermi units will still try to be put on my GTS 250. Right now, "GPU 0: FERMI:1 G92 [GeForce 9800 GTX+]" is the only GPU that works, and it's my GTX 470. "GPU 1: FERMI:1 GF100 [GeForce GTX 480]" doesn't work.
heydabop
 
Posts: 8
Joined: Tue Mar 29, 2011 11:10 pm

Re: #597 GPU Detection error: 2x Fermi should be 1xFermi + 1

Postby HendricksSA » Thu Mar 31, 2011 6:34 pm

I'm new to v7 so please allow me a quick question. Heydabop, isn't your card a 470? Why is it being called a 480?
HendricksSA
 
Posts: 545
Joined: Fri Jun 26, 2009 4:34 am

Next

Return to V7.1.52 Windows/Linux

Who is online

Users browsing this forum: No registered users and 2 guests

cron