Minor warning for anyone tempted to run every Beta test, and something for Bernd to think about when writing app_info.xml files.
You cannot switch tasks between CUDA and CPU just by preserving the version number. The is embedded in client_state.xml, and a matching needs to be in the new app_info.xml to preserve the task.
For example: I just loaded the new v3.09 Beta (CPU only) without modification on my CUDA test machine. As expected, on restart I got:
Quote:
08/08/2009 11:58:52 Einstein@Home [error] No application found for task: windows_intelx86 307 cuda; discarding
Not a problem - I was expecting it, and it was an unstarted task with no wasted computing time. But any work in progress, or completed work ready to report, would have been lost.
In any event, here at Einstein it was quickly followed by:
Quote:
08/08/2009 11:59:57 Einstein@Home Message from server: Resent lost result p2030_53835_43933_0121_G67.73-00.22.C_6.dm_302_1
and assigned to the v3.09 CPU app - which is exactly what I wanted to happen.
I have installed it on one machine running WinXP with a GTS250, nvidia drivers 190.38, cuda 2.3 DLL's and BOINC 6.6.37. It has picked up 3 ABP1 wu, the 1st of which is running as I type this. Normal crunch time for these is approx 8 hours, so I will have to report progress later.
Both units ends with message "[ERROR] Error during CUDA host->device data transfer (unknown error)". I think the problem is related to your 64bit-OS (error with Vista64 was reported earlier in this thread) and W7.
People at GPUGrid are reporting errors with the GTX260 and 190.38 drivers (one even narrowed it down to the 192 version, has a 216 which works OK). Adding W7 and x64 to the mix adds to the probability that it's a driver issue.
I have installed it on one machine running WinXP with a GTS250, nvidia drivers 190.38, cuda 2.3 DLL's and BOINC 6.6.37. It has picked up 3 ABP1 wu, the 1st of which is running as I type this. Normal crunch time for these is approx 8 hours, so I will have to report progress later.
And now: RE: 07/08/2009
)
And now:
:(
Windows 7 64 bit, BOINC
)
Windows 7 64 bit, BOINC 6.6.28, Core 2 Quad Q9400, GeForce 9400GT, driver version: 181.71
12 wu, 12 error, all with exit status 3 (0x3)
This is a wu for example: 57042735
It works, no problem, it's
)
It works, no problem, it's fine for me. :-)
Used Computer: 1935062 (Phenom II X4 940 + GTX 260 + GTX 295)
CUDA-Details on GPUGrid-Host (Driver 190.38)
Here we go: 136258596, 136258545, 136258527, 136258486, 136258460, 136258432, 136258406, 136258314, 136258249, 136226450, 136226437, 136226429, 136226399
On CPU, a WU needs average 9:40 hours. On GPU, a WU needs average 5:10 minutes. That is about 45% faster.
Einstein needed per GPU one CPU core, therefore it runs only one other CPU Task. But the cores are not always fully loaded.
There were no problems to run GPUGrid at the same time. With Einstein, the cards are not so hot.
GTX 260: Einstein: 61°/ GPUGrid 79°
GTX 295a: Einstein: 75°/ GPUGrid 88°
GTX 295b: Einstein: 73°/ GPUGrid 85°
If you need more info, please ask. I hope I can reply in German. ;-)
Bye, Grubix.
Minor warning for anyone
)
Minor warning for anyone tempted to run every Beta test, and something for Bernd to think about when writing app_info.xml files.
You cannot switch tasks between CUDA and CPU just by preserving the version number. The is embedded in client_state.xml, and a matching needs to be in the new app_info.xml to preserve the task.
For example: I just loaded the new v3.09 Beta (CPU only) without modification on my CUDA test machine. As expected, on restart I got:
Not a problem - I was expecting it, and it was an unstarted task with no wasted computing time. But any work in progress, or completed work ready to report, would have been lost.
In any event, here at Einstein it was quickly followed by:
and assigned to the v3.09 CPU app - which is exactly what I wanted to happen.
Running score so far, all on the same host:
v3.07 (CPU): ~38,000 seconds
v3.07 (CUDA): ~21,000 seconds (6 validated, 2 pending, no errors except the initial DLL)
v3.08 (CPU): ~28,000 seconds
This time, the next installment will be in the v3.09 (CPU) thread....
I have installed it on one
)
I have installed it on one machine running WinXP with a GTS250, nvidia drivers 190.38, cuda 2.3 DLL's and BOINC 6.6.37. It has picked up 3 ABP1 wu, the 1st of which is running as I type this. Normal crunch time for these is approx 8 hours, so I will have to report progress later.
Links to wu:
1st wu
2nd wu
3rd wu
BOINC blog
einsteinbinary_ABP1_3.07_wind
)
einsteinbinary_ABP1_3.07_windows_intelx86_cuda.exe exiting with error, WU processing is also interrupts:
09.08.2009 10:52:51 Einstein@Home Computation for task p2030_53995_04565_0036_G64.21-02.82.C_2.dm_560_4 finished
09.08.2009 10:52:51 Einstein@Home Output file p2030_53995_04565_0036_G64.21-02.82.C_2.dm_560_4_0 for task p2030_53995_04565_0036_G64.21-02.82.C_2.dm_560_4 absent
This and this tasks.
Win 7 x64, GTX260 (192 s.b.), driver 190.38. BOINC 6.6.37.
RE: Win 7 x64, GTX260 (192
)
Both units ends with message "[ERROR] Error during CUDA host->device data transfer (unknown error)". I think the problem is related to your 64bit-OS (error with Vista64 was reported earlier in this thread) and W7.
People at GPUGrid are
)
People at GPUGrid are reporting errors with the GTX260 and 190.38 drivers (one even narrowed it down to the 192 version, has a 216 which works OK). Adding W7 and x64 to the mix adds to the probability that it's a driver issue.
RE: I have installed it on
)
All 3 completed successfully. All ran in 5 hrs 45 mins.
Cheers,
MarkJ
BOINC blog
http://einstein.phys.uwm.edu/
)
http://einsteinathome.org/task/136430323
crashed for me on win7 64bit, geforce 285, driver 190.38
Florian
www.domplatz1.de