Problem with Quadro 600 (GF108) and Core_18 WU

[Kebast]

One of my cards drew a 0x18 WU that it will barely complete before the deadline. How can I prevent this from happening again?
Log:

Code: Select all

21:38:04:      Version: 7.4.4
21:38:04:         Date: Mar 4 2014
21:38:04:         Time: 20:26:54
21:38:04:      SVN Rev: 4130
21:38:04:       Branch: fah/trunk/client
21:38:04:     Compiler: Intel(R) C++ MSVC 1500 mode 1200
21:38:04:      Options: /TP /nologo /EHa /Qdiag-disable:4297,4103,1786,279 /Ox -arch:SSE
21:38:04:               /QaxSSE2,SSE3,SSSE3,SSE4.1,SSE4.2 /Qopenmp /Qrestrict /MT /Qmkl
21:38:04:     Platform: win32 XP
21:38:04:         Bits: 32
21:38:04:         Mode: Release
21:38:04:******************************* System ********************************
21:38:04:          CPU: Intel(R) Xeon(R) CPU E5603 @ 1.60GHz
21:38:04:       CPU ID: GenuineIntel Family 6 Model 44 Stepping 2
21:38:04:         CPUs: 8
21:38:04:       Memory: 3.50GiB
21:38:04:  Free Memory: 2.39GiB
21:38:04:      Threads: WINDOWS_THREADS
21:38:04:   OS Version: 6.1
21:38:04:  Has Battery: false
21:38:04:   On Battery: false
21:38:04:   UTC Offset: -4
21:38:04:          PID: 3496
21:38:04:          CWD: C:/Users/Mandli_06/AppData/Roaming/FAHClient
21:38:04:           OS: Windows 7 Professional Service Pack 1
21:38:04:      OS Arch: X86
21:38:04:         GPUs: 1
21:38:04:        GPU 0: NVIDIA:2 GF108 [Quadro 600]
21:38:04:         CUDA: 2.1
21:38:04:  CUDA Driver: 7050
21:38:04:Win32 Service: false
21:38:04:***********************************************************************
21:38:04:<config>
21:38:04:  <!-- Network -->
21:38:04:  <proxy v=':8080'/>
21:38:04:
21:38:04:  <!-- Slot Control -->
21:38:04:  <power v='full'/>
21:38:04:
21:38:04:  <!-- User Information -->
21:38:04:  <passkey v='********************************'/>
21:38:04:  <team v='229226'/>
21:38:04:  <user v='Kebast'/>
21:38:04:
21:38:04:  <!-- Folding Slots -->
21:38:04:  <slot id='0' type='CPU'>
21:38:04:    <cpus v='8'/>
21:38:04:  </slot>
21:38:04:  <slot id='1' type='GPU'>
21:38:04:    <client-type v='beta'/>
21:38:04:  </slot>
21:38:04:</config>
21:38:04:Trying to access database...
21:38:06:Successfully acquired database lock
21:38:06:Enabled folding slot 00: READY cpu:8
21:38:06:Enabled folding slot 01: READY gpu:0:GF108 [Quadro 600]
21:38:09:WU00:FS01:Starting
21:38:09:WU00:FS01:Running FahCore: "C:\Program Files\FAHClient/FAHCoreWrapper.exe" C:/Users/Mandli_06/AppData/Roaming/FAHClient/cores/web.stanford.edu/~pande/Win32/x86/NVIDIA/Fermi/beta/Core_18.fah/FahCore_18.exe -dir 00 -suffix 01 -version 704 -lifeline 3496 -checkpoint 15 -gpu 0 -gpu-vendor nvidia
21:38:10:WU00:FS01:Started FahCore on PID 4732
21:38:16:WU00:FS01:Core PID:4988
21:38:16:WU00:FS01:FahCore 0x18 started
21:38:17:WU01:FS00:Starting

[Joe_H]

The only assistance given in the public forum for the use of the beta flag is in combination with the max-packet-size parameter being set to the value of small as an indication you are opting into folding the older Core_15 WU's. That would keep your system from being assigned Core_17 and later GPU projects. That is probably a better fit for your GPU. Being approximately equivalent to a GT 430, it is at about the minimum to process the later projects.

Otherwise you will need to reduce your CPU count by at least one from 8, the later GPU folding cores for nVidia require one CPU core available to process and transfer data to and from the GPU.

[Kebast]

Thanks! I reduced the allocated CPU cores from 8 to 7 to hopefully finish this WU a bit faster. I'll add the small parameter as well.

[Kebast]

Oddly enough, reducing the CPU cores from 8 to 7 had zero affect on the GPU processing. It did, however, increase the production for the CPU 0xa4 WU. In any case, hopefully I've taken steps to prevent my GPU from getting these WU in the future.

[Kebast]

Only thing that seems to affect processing time is overclocking the GPU. I bumped up the shader clock a bit and it shaved a day off the total time. I was surprised to see an almost exact increase in PPD vs % overclock.