Forum > GPU crunching
V11 of CUDA MB mod - attempt to restart freezed apps
Zeus Fab3r:
--- Quote from: Raistmer on 05 Apr 2009, 07:15:27 am ---Fine, I'm happy that problem disappears w/o such watchdog version using (it will slow down processing, very small slowdown though but ...).
I use it more than week on my dual GPU host and no more hangs I noticed. Maybe it helps, maybe simple no hangups - I didn't look in every stderr to findout there were restarts or not.
No invalid results from that host - it's enough for me for now.
If someone encounter some bug - he can report it. I'm busy with AP and newly recived ATI cards now.
--- End quote ---
Finally a solution that actually works ;D
3 days, 24/7 w/o single hangup and counting. Great job !
Still have a question. Can I run this cross_watch mod alongside opt AP?
Thanks again Raistmer.
Raistmer:
--- Quote from: bbokica on 09 Apr 2009, 06:21:32 am ---Still have a question. Can I run this cross_watch mod alongside opt AP?
--- End quote ---
As long as at least 2 GPU apps running too. BOINC can't schedule correctly other apps with team MB pack. But you could try to use BOINC 6.6.20 (recommended one now) and this cross-watch build w/o CPU teampack part. Not usre what result it will give, but maybe it will work OK.
chelski:
I'm using MB_6.08_mod_CUDA_V11_def_func_FFTW_ESTIMATE_update in a V9 structure (e.g. without a number_of_cpus) in order to run AP on CPUs and MB on GPUS and still get the occasional freeze at 0%. Is there a workaround for this structure - e.g. cross watch not between CPU-MB app and GPU-MB app but between AP and GPU-MB app to solve the freezing issue?
Please correct me if I'm wrong about how this thing works... things have really moved on rather fast and it is not that clear anymore which app / build to use for which setup. Thanks
Borgholio:
Currently running the non-team V9 pack on Boinc 6.6.20 and it's working fine aside from the occasional CUDA app freeze. I have a task set under Windows to run benchmarks every half hour which frees up most stuck tasks but from time to time a task will freeze that requires a restart of BOINC. If I were to download V11, would I need to upgrade from V9 to V10 first, or could I simply extract the CUDA app from V11 and modify my app-info to look at the new file instead?
Raistmer:
--- Quote from: chelski on 18 Apr 2009, 01:36:31 pm ---I'm using MB_6.08_mod_CUDA_V11_def_func_FFTW_ESTIMATE_update in a V9 structure (e.g. without a number_of_cpus) in order to run AP on CPUs and MB on GPUS and still get the occasional freeze at 0%. Is there a workaround for this structure - e.g. cross watch not between CPU-MB app and GPU-MB app but between AP and GPU-MB app to solve the freezing issue?
Please correct me if I'm wrong about how this thing works... things have really moved on rather fast and it is not that clear anymore which app / build to use for which setup. Thanks
--- End quote ---
No, there is no such mod. But it's possible.
Navigation
[0] Message Index
[#] Next page
[*] Previous page
Go to full version