Seti@Home optimized science apps and information
Optimized Seti@Home apps => Windows => GPU crunching => Topic started by: Maik on 08 Jan 2009, 01:54:15 pm
-
found in: SETI@home community | Questions & answers | CUDA (http://setiathome.berkeley.edu/forum_forum.php?id=631)
-
Thanx for info, will give them a try too. Maybe it will solve VLAR trouble....
-
cached 20 new WU's (all medium AR). lets see ...
Edit:
- WU's need same time to crunch
- a bit less laggy
- AR: 0.3872 goes stuck at 0.000% (its fom 16no08ai, just my feeling that WU's from 16no08-series bring the trouble )
Edit2:
- did a better install of new drivers. maybe it works better
-
Hmm... new driver doesn't fix the problem:
WU true angle range is : 0.437221 (01no08ac)
SETI@Home Informational message -9 result_overflow
NOTE: The number of results detected exceeds the storage space allocated.
Flopcounter: 21739917742.860546
Spike count: 30
Pulse count: 0
Triplet count: 0
Gaussian count: 0
called boinc_finish
-
VLAR driver crash still here.
-
very bad with new driver all 10 new pack crash don't wanna start at all
-
very bad with new driver all 10 new pack crash don't wanna start at all
Most likely you just have bad bunch of tasks.
-
very bad with new driver all 10 new pack crash don't wanna start at all
Most likely you just have bad bunch of tasks.
Or the first shoots your gfx to hell. Did a reboot?
-
yes 5 times reboot same 20 pack for
now just crunching astropulse
-
i tihnk for vistas are 185.20 beta better for crunching because have device overlap implemented ....
for playing are better 181.20 of course
-
i use windows xp x64
-
There is -> http://www.laptopvideo2go.com/
On this site you can get driver for all Window-Versions, also WHQL and beta!
PS.:
And if the org. INF not include the GPU, use the modded INF ;-)
-
better is guru3d.com ...
-
The new drivers fixed my constant crashing in the Nvidia drivers, no more blue screen crashes.
Just completed 40 cuda wu without problems using the versions posted by Raistmer.
-
Ok, first WU detected that kills your gfx despite of new drivers.
using stock app 606
WU true angle range is : 0.009233
Cuda error 'cudaMemcpy(best_PoT, dev_tmp_pot, max_nb_of_elems * sizeof(float), cudaMemcpyDeviceToHost)' in file 'c:/sw/gpgpu/seti/seti_boinc/client/cuda/cudaAcc_pulsefind.cu' in line 1265 : unknown error.
full stderr attached
[attachment deleted by admin]
-
Oh WTF ?!?
First time i got a blue screen ....
All written down here. (http://setiathome.berkeley.edu/forum_thread.php?id=51283)
-
Yes, as I said already new driver doesn't fix VLAR driver crash bug. Have strong considerations that this bug connected with 2-3 seconds limit for execution time for single CUDA kernel. This limitation exists in CUDA itself - OS just restart driver when there is no respond from it for few seconds. Probably the single way to eliminate this bug for now is to split affected kernel on few CUDA calls.
-
its a watchdog problem on primary display in windows core - it happens on devices where is a primary display extended . windows core is waiting for device response in 5 seconds max intervals if nothing happens it crashed to BSOD in xp or restart devce driver in vista. so if kernel take more execution time as 5 secs this is problem for
windows core. on devices without primary display u can run kernels with any execution times ....
-
its a watchdog problem on primary display in windows core - it happens on devices where is a primary display extended . windows core is waiting for device response in 5 seconds max intervals if nothing happens it crashed to BSOD in xp or restart devce driver in vista. so if kernel take more execution time as 5 secs this is problem for
windows core. on devices without primary display u can run kernels with any execution times ....
Very likely, BUT there is too little number of such non-primarily devices available. And I personally have no one (actually have one PC with dual GeForce but it's x64 host and there are other troubles - it gives computational errors always it seems :( )
So, if you have such setup (CUDA GPU with no primarily display), please, try to redo VLAR tests from this thread: http://lunatics.kwsn.net/gpu-crunching/wus-that-cuda-mb-cant-do-correctly.0.html.
And report results. It would be helpfull for debugging I think.