GPU3 (Core 15) & Linux discussion

Moderators: slegrand, Site Moderators, PandeGroup

Re: GPU3 (Core 15) & Linux discussion

Postby swl1973 » Mon Jul 19, 2010 7:05 pm

Yes, Hyperlife's solution works -- I'm now getting Core 15 units.

But I'm not sure why it works. The DLL search path for Windows (http://msdn.microsoft.com/en-us/library/7d83bc18%28VS.71%29.aspx is:

    With both implicit and explicit linking, Windows first searches the set of pre-installed DLLs such as the performance library (KERNEL32.DLL) and the security library (USER32.DLL). Windows then searches for the DLLs in the following sequence:
    1. The directory where the executable module for the current process is located.
    2. The current directory.
    3. The Windows system directory. The GetSystemDirectory function retrieves the path of this directory.
    4. The Windows directory. The GetWindowsDirectory function retrieves the path of this directory.
    5. The directories listed in the PATH environment variable.
The DLL search path for Wine (http://www.winehq.org/docs/wineusr-guide/config-wine-main)is:

    DLLs usually get loaded in the following order:
    1. The directory the program was started from.
    2. The current directory.
    3. The Windows system directory.
    4.The Windows directory.
    5. The PATH variable directories.

Since we had the DLLs in the directory with the executable, it should have worked before.
swl1973
 
Posts: 8
Joined: Wed Jun 16, 2010 11:01 pm

Re: GPU3 (Core 15) & Linux discussion

Postby Hyperlife » Mon Jul 19, 2010 7:39 pm

My guess is that the client uses a hard-coded path to %SystemRoot%/system32/nvcuda.dll for the sole purpose of determining GPU type and CUDA version. Everything else seems to work OK when the libraries are in the same directory as the client.

I put all of the files and symlinks in the system32 directory just for simplicity's sake. You could probably leave all the files and symlinks in the client directory as long as you put a nvcuda.dll symlink in the system32 directory.
Image
User avatar
Hyperlife
 
Posts: 447
Joined: Sun Dec 02, 2007 7:38 am

Re: GPU3 (Core 15) & Linux discussion

Postby Sidicas » Tue Jul 20, 2010 1:26 am

Hyperlife wrote:My guess is that the client uses a hard-coded path to %SystemRoot%/system32/nvcuda.dll for the sole purpose of determining GPU type and CUDA version. Everything else seems to work OK when the libraries are in the same directory as the client.

/facepalm...
Glad to hear you got that part figured out... Yes, in your case you will still need the forcegpu because you're running headless and Wine throws a fit about Direct3D without DRI..

If somebody who isn't headless did the same as you just did, they could try it without the forcegpu flag and there's a good chance it will work...
Sidicas
 
Posts: 233
Joined: Sun Feb 17, 2008 4:46 pm

Re: GPU3 (Core 15) & Linux discussion

Postby swl1973 » Tue Jul 20, 2010 2:03 am

Just tried it without the "-forcegpu nvidia_g80" on both GPUs (one with an attached monitor and one without) and I got Wine popup boxes "At present your GPU is not supported or you need a current driver (etc)" So you still need the flag even if you are running X.
swl1973
 
Posts: 8
Joined: Wed Jun 16, 2010 11:01 pm

Re: GPU3 (Core 15) & Linux discussion

Postby Sidicas » Tue Jul 20, 2010 3:55 pm

swl1973 wrote:Just tried it without the "-forcegpu nvidia_g80" on both GPUs (one with an attached monitor and one without) and I got Wine popup boxes "At present your GPU is not supported or you need a current driver (etc)" So you still need the flag even if you are running X.

Wine version? Wine versions prior to 1.2rc2 reported earlier driver versions than 197.45 regardless of what nvidia driver version you have installed..

The nvidia driver version is hard-coded into Wine. http://source.winehq.org/source/dlls/wined3d/directx.c?v=wine-1.2-rc2

I believe the folding@home client is expecting Windows driver version 197.45 for Core15 workunits.
The forcegpu flag not only skips GPU detection, it also skips driver detection..
Sidicas
 
Posts: 233
Joined: Sun Feb 17, 2008 4:46 pm

Re: GPU3 (Core 15) & Linux discussion

Postby swl1973 » Wed Jul 21, 2010 6:11 am

I am running wine-1.2-rc3 which looks like it should report 197.45 for my 9600GSOs.
swl1973
 
Posts: 8
Joined: Wed Jun 16, 2010 11:01 pm

Re: GPU3 (Core 15) & Linux discussion

Postby Skiesare » Thu Jul 22, 2010 3:27 pm

I just started the client for the first time in a few weeks and picked up a core 15 WU first time, (10633 OPENMMGMX_test) I'm using 256.35 drivers with a GTX260 and using the command line "wine ./Folding@home-Win32-GPU.exe -forcegpu nvidia_g80" with wine 1.2.
Skiesare
 
Posts: 17
Joined: Mon Aug 25, 2008 6:46 am

Re: GPU3 (Core 15) & Linux discussion

Postby draeh » Wed Oct 06, 2010 6:32 pm

Sorry to resurrect such an old discussion, but I ran into a problem today when my rig finally downloaded Core 15.

Code: Select all
cudaMemcpyToSymbol: SetSim copy to cSim failed invalid device symbol


I followed the NVIDIA GPU3 Linux/Wine Headless Install Guide on Ubuntu 10.04 using an 8800GTX.

Any suggestions?
draeh
 
Posts: 16
Joined: Fri May 02, 2008 12:51 pm

Re: GPU3 (Core 15) & Linux discussion

Postby jimerickson » Wed Oct 06, 2010 6:38 pm

doesn't core 15 only run on fermi cards?
jimerickson
 
Posts: 770
Joined: Tue May 27, 2008 11:56 pm
Location: ames, iowa

Re: GPU3 (Core 15) & Linux discussion

Postby PantherX » Wed Oct 06, 2010 6:44 pm

jimerickson wrote:doesn't core 15 only run on fermi cards?
Nope, they run on Non-Fermi too. However, I am not sure if they are the same version or not.
User avatar
PantherX
 
Posts: 6614
Joined: Wed Dec 23, 2009 9:33 am

Re: GPU3 (Core 15) & Linux discussion

Postby jimerickson » Wed Oct 06, 2010 7:02 pm

thanks for the information PantherX, i stand corrected.
jimerickson
 
Posts: 770
Joined: Tue May 27, 2008 11:56 pm
Location: ames, iowa

Re: GPU3 (Core 15) & Linux discussion

Postby Hyperlife » Thu Oct 07, 2010 12:17 am

Haven't seen that error before. I haven't tested this on Ubuntu 10.04 yet.

Could you try deleting the 3.0 toolkit and installing 3.1? Delete everything in /usr/local/cuda, then download and install the 3.1 toolkit from here:

http://developer.download.nvidia.com/co ... tu9.10.run

You may need to link the new toolkit with "sudo ldconfig" after installing.
User avatar
Hyperlife
 
Posts: 447
Joined: Sun Dec 02, 2007 7:38 am

Re: GPU3 (Core 15) & Linux discussion

Postby Shelnutt2 » Thu Oct 07, 2010 12:36 pm

After you do that (^^^^) if it still doesn't work, can you post the whole terminal output, plus the output of "glxinfo" and "./usr/local/cuda/C/bin/release/deviceQuery"
OCforums, T32

Linux Lober!
Shelnutt2
 
Posts: 150
Joined: Sun Dec 02, 2007 5:37 pm

Re: GPU3 (Core 15) & Linux discussion

Postby draeh » Thu Oct 07, 2010 12:47 pm

Hyperlife wrote:Haven't seen that error before. I haven't tested this on Ubuntu 10.04 yet.

Could you try deleting the 3.0 toolkit and installing 3.1? Delete everything in /usr/local/cuda, then download and install the 3.1 toolkit from here:

http://developer.download.nvidia.com/co ... tu9.10.run

You may need to link the new toolkit with "sudo ldconfig" after installing.


Will do. I'm just waiting for my current gpu2 fold to finish.

I assume that I should also use the wrappers from here (wrapper-3.1):

Code: Select all
http://www.linuxfah.info/wrapper-3.1/cudart/cudart.dll.so
http://www.linuxfah.info/wrapper-3.1/cufft/cufft.dll.so


Shelnutt2 wrote:After you do that (^^^^) if it still doesn't work, can you post the whole terminal output, plus the output of "glxinfo" and "./usr/local/cuda/C/bin/release/deviceQuery"


Will do.

I'll report back soon.
draeh
 
Posts: 16
Joined: Fri May 02, 2008 12:51 pm

Re: GPU3 (Core 15) & Linux discussion

Postby draeh » Thu Oct 07, 2010 4:06 pm

Shelnutt2 wrote:After you do that (^^^^) if it still doesn't work, can you post the whole terminal output, plus the output of "glxinfo" and "./usr/local/cuda/C/bin/release/deviceQuery"



Same error occurs with toolkit v3.1.

glxinfo doesn't seem to want to work on a headless box.

Code: Select all
gheard@c2d:~$ glxinfo
Error: unable to open display


deviceQuery...
Code: Select all
./deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

There is 1 device supporting CUDA

Device 0: "GeForce 8800 GTX"
  CUDA Driver Version:                           3.10
  CUDA Runtime Version:                          3.10
  CUDA Capability Major revision number:         1
  CUDA Capability Minor revision number:         0
  Total amount of global memory:                 804585472 bytes
  Number of multiprocessors:                     16
  Number of cores:                               128
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       16384 bytes
  Total number of registers available per block: 8192
  Warp size:                                     32
  Maximum number of threads per block:           512
  Maximum sizes of each dimension of a block:    512 x 512 x 64
  Maximum sizes of each dimension of a grid:     65535 x 65535 x 1
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             256 bytes
  Clock rate:                                    1.35 GHz
  Concurrent copy and execution:                 No
  Run time limit on kernels:                     No
  Integrated:                                    No
  Support host page-locked memory mapping:       No
  Compute mode:                                  Default (multiple host threads can use this device simultaneously)
  Concurrent kernel execution:                   No
  Device has ECC support enabled:                No

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 3.10, CUDA Runtime Version = 3.10, NumDevs = 1, Device = GeForce 8800 GTX


PASSED

Press <Enter> to Quit...
-----------------------------------------------------------


Folding log...
Code: Select all

Note: Please read the license agreement (Folding@home-Win32-GPU.exe -license). Further
use of this software requires that you have read and accepted this agreement.

[14:18:36] Gpu type=2 species=31.


--- Opening Log file [October 7 14:18:36 UTC]


# Windows GPU Console Edition #################################################
###############################################################################

                       Folding@Home Client Version 6.30r1

                          http://folding.stanford.edu

###############################################################################
###############################################################################

Launch directory: Z:\home\gheard\fahgpu3
Executable: Z:\home\gheard\fahgpu3\Folding@home-Win32-GPU.exe
Arguments: -verbosity 9 -forcegpu nvidia_g80 -pause

[14:18:36] - Ask before connecting: No
[14:18:36] - User name: draeh (Team 78445)
[14:18:36] - User ID: 7A8D75740383D6FA
[14:18:36] - Machine ID: 1
[14:18:36]
[14:18:36] Gpu type=2 species=31.
[14:18:37] Loaded queue successfully.
[14:18:37]
[14:18:37] - Autosending finished units... [14:18:37]
[14:18:37] + Processing work unit
[14:18:37] Trying to send all finished work units
[14:18:37] Core required: FahCore_15.exe
[14:18:37] + No unsent completed units remaining.
[14:18:37] - Autosend completed
[14:18:37] Core found.
[14:18:37] Working on queue slot 03 [October 7 14:18:37 UTC]
[14:18:37] + Working ...
[14:18:37] - Calling '.\FahCore_15.exe -dir work/ -suffix 03 -nice 19 -checkpoint 5 -verbose -lifeline 8 -version 630'

Warning: Ignoring unknown arg
Warning: Ignoring unknown arg
[14:18:37]
[14:18:37] *------------------------------*
[14:18:37] Folding@Home GPU Core -- Beta
[14:18:37] Version 2.09 (Thu May 20 11:51:02 PDT 2010)
[14:18:37]
[14:18:37] Compiler  : Microsoft (R) 32-bit C/C++ Optimizing Compiler Version 14.00.50727.42 for 80x86
[14:18:37] Build host: amoeba
[14:18:37] Board Type: Nvidia
[14:18:37] Core      :
[14:18:37] Preparing to commence simulation
[14:18:37] - Ensuring status. Please wait.
[14:18:46] - Looking at optimizations...
[14:18:46] - Working with standard loops on this execution.
[14:18:46] Examination of work files indicates 8 consecutive improper terminations of core.
[14:18:46] sizeof(CORE_PACKET_HDR) = 512 file=<>
[14:18:46] - Expanded 43770 -> 163067 (decompressed 372.5 percent)
[14:18:46] Called DecompressByteArray: compressed_data_size=43770 data_size=163067, decompressed_data_size=163067 diff=0
[14:18:46] - Digital signature verified
[14:18:46]
[14:18:46] Project: 10632 (Run 43, Clone 93, Gen 6)
[14:18:46]
[14:18:46] Entering M.D.
[14:18:52] Tpr hash work/wudata_03.tpr:  3979973129 712644698 2390509216 2220500147 2677046037
[14:18:52] Working on 582 p2750_N68H_AM03
[14:18:52] Client config found, loading data.
Reading file work/wudata_03.tpr, VERSION 4.0.5 (single precision)
Note: tpx file_version 58, software version 59
[14:18:53] Starting GUI Server
cudaMemcpyToSymbol: SetSim copy to cSim failed invalid device symbol
Run: exception thrown during GuardedRun
[14:18:53] Run: exception thrown during GuardedRun
[14:18:53] Run: exception thrown in GuardedRun -- Gromacs cannot continue further.
[14:18:53] Going to send back what have done -- stepsTotalG=0
[14:18:53] Work fraction=0.0000 steps=0.
[14:18:57] logfile size=7228 infoLength=7228 edr=0 trr=23
[14:18:57] + Opened results file
[14:18:57] - Writing 7764 bytes of core data to disk...
[14:18:57] Done: 7252 -> 2485 (compressed to 34.2 percent)
[14:18:57]   ... Done.
[14:18:57] DeleteFrameFiles: successfully deleted file=work/wudata_03.ckp
[14:18:57]
[14:18:57] Folding@home Core Shutdown: UNSTABLE_MACHINE
draeh
 
Posts: 16
Joined: Fri May 02, 2008 12:51 pm

PreviousNext

Return to unOfficial Linux GPU (WINE wrapper) (3rd party support)

Who is online

Users browsing this forum: No registered users and 1 guest

cron