Posts by Tomohisa Miya

1) Message boards : Number crunching : Wierd number on 'applications' page. (Message 62828) Posted 12 Dec 2014 by Tomohisa Miya Post: What happen? The applications page says, average computing of nVIDIA OpenCL apps for Mac is over 26 ZetaFLOPS!
2) Message boards : Application Code Discussion : MAC OS X Nvidia - MilkyWay Sepearation Modified Fit v1.30 (Message 62810) Posted 9 Dec 2014 by Tomohisa Miya Post: Just fort memo. Difference of OpenCL configuration for multiple type of GPUs @ Windows : OpenCL drivers are provided by each GPU vender. Platform #0 VENDER-A - device_index 0 GPU-A #0 - device_index 1 GPU-A #1 : Platform #1 VENDER-B - device_index 0 GPU-B #0 - device_index 1 GPU-B #1 : So, the device index matches with device number. @ Mac OS X : OpenCL driver is provided by Apple as standard feature of OS Platform #0 Apple - device_index 1 GPU-A #0 : - device_index N-1 GPU-A #N - device_index N GPU-B #0 : So, the device index won't match with device number for GPU-Bs The boinc_get_opencl_ids() tries to get correct platform ID and device ID by scanning all OpenCL devices and counting only GPUs of the specifeied verdor.
3) Message boards : Number crunching : Computation ERROR (Message 62809) Posted 9 Dec 2014 by Tomohisa Miya Post: Hi Phil, I'm also having the same problem as you. # my mac looks identical as yours, except OS version. Unfortunately, the use of <exclude_gpu> doesn't work, at least for me. It's known problem and on discussion at the Application Code Discussion board http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=3534. Seemingly, it is a problem in the MW application, which doesn't expected the way Apple implemented OpenCL for multiple type of GPUs on Mac. As far as I know, some other projects can run OpenCL apps properly on the mac. So, I hope it won't take long to address the issue. I seem to remember that at one time there was a configuration parameter that could be set to prevent using the GPU. Maybe I can find that to turn these WU's off. I did below to make MW not asking for NVIDIA tasks. - On BOINC Manager of the mac, 1 Set [No new tasks] for MW - From your account page on this site, 2. Set the mac's [location] to which others are not using. ( in [Details] of the computer ) 3. Uncheck [Use NVIDIA GPU] for the location. ( in [MilkyWay@Home preferences] ) # Changes become active after [update] - On BOINC Manager of the mac, 4. Do [Update] for MW and wait a while to complete the request. 5. Do [Allow new task] for MW. To make it back, - Change the location back to the original, or check the [Use NVIDIA GPU] again. - Then do step 4 above. Hoping this will be a help, Tommy
4) Message boards : Application Code Discussion : MAC OS X Nvidia - MilkyWay Sepearation Modified Fit v1.30 (Message 62793) Posted 6 Dec 2014 by Tomohisa Miya Post: Correction: I found the below while running other BOINC project that has OpenCL applications availablee for both of GPUs, nVIDIA and Intel GPU. Those two application run concurrently without problem even both get the same command line option --device 0. I guess the device number 0 means the first GPU of that kind, not the first one of all kind of GPU. So, skipping other type of OpenCL GPU may make the problem gone. # I still believe the use of boinc_get_opencl_ids() is better. Because it is the way written in the BOINC document. Regards, Tomohisa Miya
5) Message boards : Application Code Discussion : MAC OS X Nvidia - MilkyWay Sepearation Modified Fit v1.30 (Message 62747) Posted 27 Nov 2014 by Tomohisa Miya Post: Hello, Jake, I got the same problem on my MacBook Pro (w/ Intel HD 4000 & nVIDIA 650M). I believe the cause is in the code of milkyway@home. The task was assigned to 650M(dev#1), which supports DP, but it failed due to the lack of DP on HD 4000(dev#0). Seemingly milkyway app failed to get proper device number of the assigned GPU. According to the document at BOINC site <http://boinc.berkeley.edu/trac/wiki/OpenclApps>, the recent BOINC clients specifies the OpenCL data in the init_data.xml file, not by command line, and the application must call boinc_get_opencl_ids(). However I cannot find it called in the milkyway source. Hoping this can be a help to address the issue. Tomohisa Miya Part of init_data.xml for the task (other project) assigned to NVIDIA GPU <gpu_type>NVIDIA</gpu_type> <gpu_device_num>0</gpu_device_num> <gpu_opencl_dev_index>1</gpu_opencl_dev_index> <gpu_usage>1.000000</gpu_usage> the stderr output of failed separation task for NVIDIA gpu <core_client_version>7.2.42</core_client_version> <![CDATA[ <message> process exited with code 1 (0x1, -255) </message> <stderr_txt> <search_application> milkyway_separation 1.36 Darwin x86_64 double OpenCL </search_application> Reading preferences ended prematurely BOINC GPU type suggests using OpenCL vendor 'NVIDIA Corporation' Setting process priority to 0 (13): Permission denied Error loading Lua script 'astronomy_parameters.txt': [string "number_parameters: 4..."]:1: '<name>' expected near '4' Switching to Parameter File Using SSE4.1 path Found 1 platform Platform 0 information: Name: Apple Version: OpenCL 1.2 (Sep 20 2014 22:01:02) Vendor: Apple Extensions: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event Profile: FULL_PROFILE Didn't find preferred platform Using device 0 on platform 0 Found 2 CL devices Device 'HD Graphics 4000' (Intel:0x1024400) (CL_DEVICE_TYPE_GPU) Board: Driver version: 1.2(Sep 25 2014 22:26:04) Version: OpenCL 1.2 Compute capability: 0.0 Max compute units: 16 Clock frequency: 1250 Mhz Global mem size: 1073741824 Local mem size: 65536 Max const buf size: 65536 Double extension: (none) Device doesn't support double precision Failed to calculate likelihood <background_integral> nan </background_integral> <stream_integral> nan nan nan </stream_integral> <background_likelihood> nan </background_likelihood> <stream_only_likelihood> nan nan nan </stream_only_likelihood> <search_likelihood> nan </search_likelihood> 01:34:25 (53761): called boinc_finish