Hi Richard, Thank you. Just to clarify, as I just posted to TBar on the SETI Beta forum: > I need the init_data.xml file information from the slot directory of a task > which has the Invalid OpenCL GPU index problem, captured before that task > finishes; those from other slots won't help.
I assume you already understood that, but it never hurts to make sure. > <warning>NVIDIA library reports 2 GPUs</warning> This is really a message for debugging; it's not really a "warning" despite what it says! It would be helpful to confirm that this app uses the API of boinc_get_opencl_ids() which takes 5 arguments, as described in <http://boinc.berkeley.edu/trac/wiki/OpenclApps>. Is the source code for this build of SETI@home beta available somewhere I can examine it? Cheers, --Charlie -- Charlie Fenton [email protected] BOINC / SETI@home Macintosh & Windows Programmer Space Sciences Laboratory UC Berkeley On Sep 18, 2014, at 2:29 AM, Richard Haselgrove <[email protected]> wrote: > I can start you off with some of that now. > > OpenCL detection: > > 16-Sep-2014 19:35:29 [---] Starting BOINC client version 7.4.21 for > windows_x86_64 > 16-Sep-2014 19:35:29 [---] log flags: file_xfer, sched_ops, task, cpu_sched, > sched_op_debug, work_fetch_debug > 16-Sep-2014 19:35:29 [---] Libraries: libcurl/7.33.0 OpenSSL/1.0.1h zlib/1.2.8 > 16-Sep-2014 19:35:29 [---] Data directory: D:\BOINCdata > 16-Sep-2014 19:35:29 [---] Running under account xxxx > 16-Sep-2014 19:35:29 [---] CUDA: NVIDIA GPU 0: GeForce GTX 670 (driver > version 337.88, CUDA version 6.0, compute capability 3.0, 2048MB, 1950MB > available, 2915 GFLOPS peak) > 16-Sep-2014 19:35:29 [---] CUDA: NVIDIA GPU 1: GeForce GTX 670 (driver > version 337.88, CUDA version 6.0, compute capability 3.0, 2048MB, 1958MB > available, 2915 GFLOPS peak) > 16-Sep-2014 19:35:29 [---] OpenCL: NVIDIA GPU 0: GeForce GTX 670 (driver > version 337.88, device version OpenCL 1.1 CUDA, 2048MB, 1950MB available, > 2915 GFLOPS peak) > 16-Sep-2014 19:35:29 [---] OpenCL: NVIDIA GPU 1: GeForce GTX 670 (driver > version 337.88, device version OpenCL 1.1 CUDA, 2048MB, 1958MB available, > 2915 GFLOPS peak) > 16-Sep-2014 19:35:29 [---] OpenCL: Intel GPU 0: Intel(R) HD Graphics 4000 > (driver version 10.18.10.3621, device version OpenCL 1.2, 990MB, 990MB > available, 154 GFLOPS peak) > 16-Sep-2014 19:35:29 [---] OpenCL CPU: Intel(R) Core(TM) i7-3770K CPU @ > 3.50GHz (OpenCL driver vendor: Intel(R) Corporation, driver version > 3.0.1.10878, device version OpenCL 1.2 (Build 76413)) > > I'd forgotten I loaded a CPU driver too! > > init_data.xml will have to follow when the GPUGrid task has finished. > > There is no app_info.xml file in this case - I'm not running anonymous > platform while Beta testing. So we can exclude that theory. Likewise, no > app_config.xml file for the project either. > > There's no <coproc> specification. The only non-standard GPU entry in > cc_config.xml is > > <exclude_gpu> > <url>http://www.gpugrid.net/</url> > <device_num>0</device_num> > <type>NVIDIA</type> > </exclude_gpu> > > - restricting GPUGrid to device 1 > > I attach coproc_info.xml, datestamped for the same startup as the log > messages above. I see > > <warning>NVIDIA library reports 2 GPUs</warning> > > - which is absolutely true, I paid for both and installed them myself! > > You'll have to ask Raistmer about the code which generates the 'wrong > platform' warning - that's not my department. I think it's unlikely to be > ATI-related, but might be Intel-related. I'll have a better idea when I can > explore more fully this afternoon. > > More to follow. > > From: Charlie Fenton <[email protected]> > To: Richard Haselgrove <[email protected]> > Cc: Raistmer the Sorcerer <[email protected]>; boinc_dev email List > <[email protected]> > Sent: Thursday, September 18, 2014 9:39 AM > Subject: Re: [boinc_dev] boinc_get_opencl_ids() returns -33 while own app > enumeration found device > > Hi Richard, > > Please send me the following when you see this problem again: > > * What does the BOINC client report about its detection of GPUs near the > beginning of BOINC's Event Log (in stdoutdae.txt a few lines after "Starting > BOINC client version 7.2.42 ....")? > > * The init_data.xml file from the slot directory with the problem. This is > the most important thing. > > * The app_info.xml file. > > * The <coproc> specification in cc_config.xml, if there is one. > > Do you know how the following message is generated? > > WARNING: BOINC supplied wrong platform! > Could this indicate that it is trying to run ATI GPU 1 instead of NVIDIA GPU > 1? > > Cheers, > --Charlie > > On Sep 18, 2014, at 1:05 AM, Richard Haselgrove > <[email protected]> wrote: > > > I'be just noticed that one of my machines is generating the same error > > messages, currently running BOINC v7.4.21 > > > > http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=61440 > > > > Machine has two identical NVidia GPUs - so uses both cards without need of > > an entry in cc_config.xml > > It also has an Intel HD 4000 iGPU, also configured for BOINC to use. > > > > I see the errors and warnings when a task is assigned to run on NV Device 1: > > Running on device number: 1 > > Priority of worker thread raised successfully > > Priority of process adjusted successfully, below normal priority class used > > Invalid OpenCL GPU index: 1 > > WARNING: boinc_get_opencl_ids failed with code -33 > > OpenCL platform detected: Intel(R) Corporation > > OpenCL platform detected: NVIDIA Corporation > > WARNING: BOINC supplied wrong platform! > > BOINC assigns device 1 > > WARNING: BOINC failed to provide OpenCL device, using own enumeration > > abilities > > > > but not when the same application is assigned to run on NV Device 0: > > Running on device number: 0 > > Priority of worker thread raised successfully > > Priority of process adjusted successfully, below normal priority class used > > OpenCL platform detected: Intel(R) Corporation > > OpenCL platform detected: NVIDIA Corporation > > BOINC assigns device 0 > > Info: BOINC provided OpenCL device ID used > > > > I normally run applications from two different projects on the two NV > > cards, which I why I haven't seen this before - and in fact I've just > > started a new task on Device 1, so it will be busy for the next 8 hours or > > so. But once it's finished, I will force SETI Beta to run on both cards, > > and forward the contrasting files for inspection. > > > > From: Raistmer the Sorcerer <[email protected]> > > To: Charlie Fenton <[email protected]> > > Cc: boinc_dev email List <[email protected]> > > Sent: Monday, September 15, 2014 6:17 PM > > Subject: Re: [boinc_dev] boinc_get_opencl_ids() returns -33 while own app > > enumeration found device > > > > Hi Charlie > > > > Please look this message: > > http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2182&postid=52412 > > > > From it one can infer that BOINC detected both GPUs and both GPUs (ATi ones > > I mean) is active, use all GPUs switch enabled. > > > > Regarding platform warning - it means that app own enumeration scheme > > detected different platform than proposed by BOINC. > > Surely it will be cause BOINC API call resulted in error. > > > > Does -33 error code corresponds OpenCL specification? If so, it probably > > means BOINC API made OpenCL 1.1 call perhaps while device is OpenCL 1.0. > > Please check this possibility. > > > > Regarding using NV instead of ATi - hardly possible. App runs on ATi GPU > > after all (and exactly on HD4xxx GPU, device 1 by means of own enumeration > > scheme). > > > > wbr > > > > > > > > > > Mon, 15 Sep 2014 05:18:08 -0700 от Charlie Fenton > > <[email protected]>: > > >Hi Raistmer, > > > > > >boinc_get_opencl_ids() reported the reason for the failure in this line: > > >> Invalid OpenCL GPU index: 1 > > >This error will occur if the value of <gpu_opencl_dev_index> provided by > > >the init_data.xml file > > > > > >It would be very helpful to see the init_data.xml file to understand what > > >went wrong. > > > > > >Does user TBar have the following option set in his cc_config.xml file? > > >> <use_all_gpus>1</use_all_gpus> > > > > > >If not, then BOINC will normally use only the most powerful ATI GPU (the > > >6770 Juniper), so the 4670 (RV730) will be ignored, so the highest valid > > >OpenCL GPU index will be 0. However, I'm not sure whether this still > > >applies in the case of anonymous platform. Also, boinc_get_opencl_ids() > > >determines the number of OpenCL devices for each platform independently. > > > > > >What does the BOINC client report about its detection of GPUs near the > > >beginning of BOINC's Event Log (in stdoutdae.txt a few lines after > > >"Starting BOINC client version 7.2.42 ....")? Does it say that the 4670 > > >is "not used"? > > > > > >The host system has one NVIDIA GPU and 2 ATI GPUs. What does this message > > >mean: > > >> WARNING: BOINC supplied wrong platform! > > > > > >Is there any possibility that the anonymous platform specification was > > >trying to run the application on a second NVIDIA GPU rather than the > > >second ATI GPU? > > > > > >Cheers, > > >--Charlie > > > > > >On Sep 14, 2014, at 12:38 AM, Raistmer the Sorcerer < [email protected] > > > >wrote: > > > > > >> Please look this post for background: > > >> > > >> > > >> http://setiweb.ssl.berkeley.edu/beta/forum_thread.php?id=2182&postid=52387 > > >> > > >> On ATI 4670 card under WinXP boinc_get_opencl_ids() returns -33. > > >> If this error code corresponds OpenCL standart it would mean > > >> #define CL_INVALID_DEVICE -33 > > >> > > >> Nevetheless app's own device enumeration abilities allow to find this > > >> GPU and use it. This results in warning given in stderr. Some another > > >> app could not work at all on such GPU relying only on BOINC enumeration > > >> scheme. > > >> Why BOINC's code fails to detect GPU correctly? > > >> > > >> - Raistmer the Sorcerer > > >> _______________________________________________ > > >> 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. > > > > > <coproc_info.zip> _______________________________________________ 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.
