GPU throttling can be effectively done via same CPU throttling as for CPU apps. Suspend app process execution for some ms every some ms (tunable option) and GPU will cool down. But what is required - to set _separate_ options in UI for CPU and GPU throttling and not to suspend processes of GPU apps when only CPU throttling active.
Четверг, 17 января 2013, 15:49 -05:00 от David Anderson <[email protected]>: >Notes: > >- When jobs are suspended for CPU throttling, they are always left in memory. >- Applications generally checkpoint at most once every few minutes. > This time scale is not useful for limiting CPU or GPU temperature. > >-- David > >On 17-Jan-2013 3:45 PM, Carlo Rigamonti wrote: >> Why not try to modify Boincmanager in order to stop the GPU at a checkpoint >> and restart after a calculated interval of time (calculated by percentage >> chosen by the user) ? >> >> I'm working on a software which try to execute the actions as above >> specified. >> >> This is an external software that check the modifications of some important >> files (init_data.xml and boinc_task_state.xml). If no checkpoint file, the >> software check only init_data.xml). >> >> Can this be a good idea ?? >> >> -----Messaggio originale----- >> Da: [email protected] >> [mailto:[email protected]] Per conto di Ed A >> Inviato: giovedì 17 gennaio 2013 17.32 >> A: S Ross >> Cc: Charles Elliott; BOINC Developers Mailing List; BOINC Alpha list >> Oggetto: Re: [boinc_alpha] [boinc_dev] "anonymous GPU" feature >> >>> OK, I now get why you want to use this - to prevent disasters, rather >> than control temperatures and performance. >> >>> Does polling the GPU cause problems with OpenCL apps as is the case with >> CPU-Z and a few similar monitoring tools? >> >> No problem with running TThrottle and any OpenCL app I've tried. >> >> Ed >> >> >> On Thu, Jan 17, 2013 at 9:57 AM, S Ross < [email protected] > wrote: >> >>> OK, I now get why you want to use this - to prevent disasters, rather than >>> control temperatures and performance. >>> >>> Modern CPU's and GPU's can automatically downclock when they are too hot. >>> Obviously if you overclock and disable fan warnings you are over-riding >>> existing emergency settings and TThrottle might help. However, in my >>> experience the most common hardware failures would be motherboard, PSU and >>> hard drive. While we don't want any catastrophic events, including GPU >>> failures, shouldn't these be the priority? If your PSU fails you can get >>> another, but if your hard drive fails you lose hardware and more >>> importantly data. >>> >>> Does polling the GPU cause problems with OpenCL apps as is the case with >>> CPU-Z and a few similar monitoring tools? >>> >>> ------------------------------ >>> *From:* Ed A < [email protected] > >>> *To:* S Ross < [email protected] > >>> *Cc:* "McLeod, John" < [email protected] >; Charles Elliott < >>> [email protected] >; BOINC Developers Mailing List < >>> [email protected] >; BOINC Alpha list < >>> [email protected] > >>> *Sent:* Thursday, 17 January 2013, 14:42 >>> >>> *Subject:* Re: [boinc_alpha] [boinc_dev] "anonymous GPU" feature >>> >>>> Using a throttling tool is the wrong solution. >>>> You can better control the GPU's temperature using tools such as EVGA >>> Precision or MSI Afterburner. >>> >>> AFAIK we were referring to catastrophic conditions. MSI Afterburner is >>> excellent for controlling fan speed and thus temps under normal >> conditions. >>> The EVGA tool only works for a very limited number of GPUs. TThrottle >>> is a latter line of defense in case of some fan failure or extremely hot >>> weather. It works. >>> >>> Regards/Ed >>> >>> >>> Adding a case fan or two might also help! >>> On Thu, Jan 17, 2013 at 8:21 AM, S Ross < [email protected] > wrote: >>> >>> Using a throttling tool is the wrong solution. >>> >>> You can better control the GPU's temperature using tools such as EVGA >>> Precision or MSI Afterburner. >>> >>> >>> On-off, on-off GPU crunching is Extremely Bad for performance; tasks >>> revert to their last checkpoint and the GPU clock will be increased and >>> decreased repeatedly (unless is stalls). >>> >>> If projects run too hot, increase the fan speed, reduce the Memory >>> frequency, downclock the GPU and/or the CPU. >>> You can even under Volt the CPU and GPU. >>> >>> TThrottle is also a poor solution to CPU temperature control, for the same >>> reasons. It's better to adjust the number of cores to crunch on, the CPU's >>> clock speeds and Voltages. >>> >>> Adding a case fan or two might also help! >>> >>> >>> >>> >>> >>> >> _______________________________________________ >> boinc_alpha mailing list >> [email protected] >> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_alpha >> To unsubscribe, visit the above URL and >> (near bottom of page) enter your email address. >> >> _______________________________________________ >> boinc_alpha mailing list >> [email protected] >> http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_alpha >> To unsubscribe, visit the above URL and >> (near bottom of page) enter your email address. >> >_______________________________________________ >boinc_dev mailing list >[email protected] >http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev >To unsubscribe, visit the above URL and >(near bottom of page) enter your email address. _______________________________________________ boinc_dev mailing list [email protected] http://lists.ssl.berkeley.edu/mailman/listinfo/boinc_dev To unsubscribe, visit the above URL and (near bottom of page) enter your email address.
