V10/11 of modified SETI MB CUDA + opt AP package for full multi-GPU+CPU use

[Richard Haselgrove]

I believe Raistmer's development system requires drivers 180.xx and above, but the stock apps work with 179.xx drivers

[xbill]

Bummer, nvidia only offers 179 for laptops. Ok, i will try to install the regular drivers (inf modding as necessary ) and test again this evening when my cache is empty.
Thanks for the tip Richard!

[rpont]

Am I understanding things correctly about the VLAR in that V10 throws them away?

If that is the case it seems a shame since my system can complete them in about 100 minutes, 4 times as long as non VLAR ones. Again if it is the case would it be possible to make killing them optional?

[The Grinch]

Driver for Notebooks > 179 see http://www.laptopvideo2go.com/

[xbill]

I did ssome more tests after i got nvidia drivers 182.06 installed (do not trust the premodded inf files from laptopvideo2go...do it yourself if you have to  )
I set up a standalone test with the short WU's
Using V10: executing MB_6.08_mod_CUDA_V10.exe works as expected:

Code: [Select]

Can't open init data file - running in standalone mode
Can't open init data file - running in standalone mode
setiathome_CUDA: Found 1 CUDA device(s):
   Device 1 : GeForce 8600M GT
           totalGlobalMem = 268435456
           sharedMemPerBlock = 16384
           regsPerBlock = 8192
           warpSize = 32
           memPitch = 262144
           maxThreadsPerBlock = 512
           clockRate = 0
           totalConstMem = 65536
           major = 1
           minor = 1
           textureAlignment = 256
           deviceOverlap = 0
           multiProcessorCount = 4
setiathome_CUDA: No device specified, determined to use CUDA device 1: GeForce 8600M GT
SETI@home using CUDA accelerated device GeForce 8600M GT
V10 modification by Raistmer
Priority of worker thread rised successfully
Priority of process adjusted successfully
Affinity of worker thread adjusted successfully
Total GPU memory 268435456 free GPU memory 176422912
setiathome_enhanced 6.02 Visual Studio/Microsoft C++

Build features: CPU-GPU team mod FFTW x86     CPUID: Intel(R) Core(TM)2 Duo CPU     T9300  @ 2.50GHz

     Cache: L1=64K L2=6144K

CPU features: FPU TSC PAE CMPXCHG8B APIC SYSENTER MTRR CMOV/CCMP MMX FXSAVE/FXRSTOR SSE SSE2 HT SSE3
libboinc: 6.1.8

Work Unit Info:
...............
WU true angle range is :  0.604884
Optimal function choices:
-----------------------------------------------------
name                
-----------------------------------------------------
              v_BaseLineSmooth (no other)
            v_GetPowerSpectrum 0.00022 0.00000
                   v_ChirpData 0.12565 0.00000
                  v_Transpose4 0.00442 0.00000
               FPU opt folding 0.00304 0.00000

Flopcounter: 745980599434.345090

Spike count:    0
Pulse count:    0
Triplet count:  0
Gaussian count: 0

Wall-clock time elapsed since last restart: 298.0 seconds
called boinc_finish

But starting the *GPU_CPU_team_V10.exe results in an application error popup from windows, no further output on the commandline, no stderr.txt. eventvwr is not helpful either: Fehlerhafte Anwendung AK_v8b_win_SSSE3x_GPU_CPU_team_V10.exe, Version 0.0.0.0, Zeitstempel 0x49a516a3, fehlerhaftes Modul AK_v8b_win_SSSE3x_GPU_CPU_team_V10.exe, Version 0.0.0.0, Zeitstempel 0x49a516a3, Ausnahmecode 0xc000000d, Fehleroffset 0x000b2d9e, Prozess-ID 0xbf8, Anwendungsstartzeit 01c998de0f35a5cf. (sorry, german locale...)
So errorcode 0xc000000d is invalid argument, that is all i can read without installing IDApro

Guess i'll try V9 for now, that seems to run without errors in standalone, hopefully it will run under BOINC as well.

Edit::

Driver for Notebooks > 179 see http://www.laptopvideo2go.com/

I could not download any driver: "We have recognized that you are attempting to directly link to our drivers." *BS* i did not filter the referer...i could download the premodded infs, but they did not work with nvidias stock drivers because they reference some files that are not contained in nvidias distribution.

[The Grinch]

Don't know where are the Problem is .. may be the Browser?!? (try to use IE6/7 - i know, IE is not the favorit...)
I'm using the drivers from laptopvideo2go sometimes without any problems, only when playback BlueRay (CopyProteced) WHQL-driver is needed.

[Jason00]

I've got a problem with my V10a x64 Cuda app, I believe it did it with the V10a x86 app too. Here's an example WU. http://setiathome.berkeley.edu/result.php?resultid=1173510764.

The message I get is this.
<message>
too many exit(0)s
</message>


Any idea on this error? It's not like I'm doing much else on this PC.

[Raistmer]

But starting the *GPU_CPU_team_V10.exe results in an application error popup from windows, no further output on the commandline, no stderr.txt. eventvwr is not helpful either: Fehlerhafte  with nvidias stock drivers because they reference some files that are not contained in nvidias distribution.

Did you start it from command line?
Team builds don't support standalone work. The should run only under BOINC and only in SETI@home main project (hard link to particular layout for CUDA MB relatively to working path required).

[Raistmer]

I've got a problem with my V10a x64 Cuda app, I believe it did it with the V10a x86 app too. Here's an example WU. http://setiathome.berkeley.edu/result.php?resultid=1173510764.

The message I get is this.
<message>
too many exit(0)s
</message>


Any idea on this error? It's not like I'm doing much else on this PC.

Build features: Non-graphics   CPU-GPU team mod   IPP   USE_SSSE3   x86      
It's x86 build, not x64 one.
Update your config to V10a. V10 had rescheduling bug that could lead to such situation. So update REQUIRED.

[Jason00]

I've got a problem with my V10a x64 Cuda app, I believe it did it with the V10a x86 app too. Here's an example WU. http://setiathome.berkeley.edu/result.php?resultid=1173510764.

The message I get is this.
<message>
too many exit(0)s
</message>


Any idea on this error? It's not like I'm doing much else on this PC.

Build features: Non-graphics   CPU-GPU team mod   IPP   USE_SSSE3   x86      
It's x86 build, not x64 one.
Update your config to V10a. V10 had rescheduling bug that could lead to such situation. So update REQUIRED.

I started using the V10a app as soon as you posted them the other day. And last night I started with the V10a Intel x64 app.

After looking at my completed WU's it shows that for my CPU WU's its x64, and the GPU WU's it's x86.

CPU
http://setiathome.berkeley.edu/result.php?resultid=1173515610
GPU
http://setiathome.berkeley.edu/result.php?resultid=1173515613

Here are the tasks for my computer.
http://setiathome.berkeley.edu/results.php?hostid=4706608

[Raistmer]

Show me x64 result with too many exits.
Result you posted before from V10 pack.

[Raistmer]

SSE3 x64 Intel-specific pack added.


[attachment deleted by admin]

[Archangel999]

Raistmer Thank YOU MAN !!! all working fine
using x64 xp pro and q6600 8800gtx all working nice and clear thank you ones again for your help


Best Regards
D.Draganov

[elec999]

When doing a Q6600 (4 cores and two gpus) should my cconfig say for cpu 6.
Thank you

[Jason G]

When doing a Q6600 (4 cores and two gpus) should my cconfig say for cpu 6.
Thank you

Since Raistmer's probably asleep I'll answer as best I can.    My understanding is 5 (because a GPU doesn't require a whole core).  So that 'extra pretend' core should accommodate many GPUs.  I hope, before too long, that  Boinc handling of this improves to the point that these cc_config workarounds become obsolete though.  Personally, deopending on what Boinc version you're running, I'd try it without the entry first, to see what happens, then up it by 1 and observe the effect.  My p4 seems to work with v9 out of the box, while my dual core doesn't... go figure!

Jason