Posts by Iain Bethune

1) Message boards : Questions and problems : BOINC does not recognise ATI GPU on Mac OS X (Message 40250) Posted 19 Sep 2011 by Iain Bethune Post: As I wrote earlier, I'm afraid I can't do much to investigate this until we get an ATI GPU to test with. I have not heard when that will be. Sorry - I'm not intending to hassle you! I do have an NVIDIA GPU, and I have now successfully run several tasks using the PPS (Sieve) 1.39 (cuda32), and have some waiting to run using the cullen/Wodall (Sieve) 1.12 (cuda32) application. Are these OpenCL or CUDA apps? The existing Mac GPU apps you mentioned are both CUDA apps. We don't currently have any OpenCL apps except the new one we are discussing PPS (Sieve) 1.39 (ati13ati). This app will run OK on Nvidia GPUs if you run it outside BOINC, but we don't plan to offer it to Nvidia clients via BOINC since the CUDA app is better optimised for Nvidia than the new OpenCL app. Cheers - Iain
2) Message boards : Questions and problems : BOINC does not recognise ATI GPU on Mac OS X (Message 40245) Posted 19 Sep 2011 by Iain Bethune Post: For now, I will just say that OpenCL applications should link with api/boinc_opencl.cpp and call thhe boinc_get_opencl_ids() API to get the cl_device_id and cl_platform_id of the GPU assigned to the application instance by BOINC's scheduler. Right now we don't do this, just using the first found platform and device, which would clearly cause problems with multiple GPU systems, and we should fix this, but at the moment it seems like the problem is not with the app, but rather with BOINC which is not even detecting the GPU correctly... The machine linked earlier is not a laptop, so the GPU switching should not be an issue. Cheers - Iain
3) Message boards : Questions and problems : BOINC does not recognise ATI GPU on Mac OS X (Message 40209) Posted 16 Sep 2011 by Iain Bethune Post: Hi, we have just released an OpenCL app for Mac over at Primegrid (see http://www.primegrid.com/forum_thread.php?id=2837#40344 for discussion). The app has been tested on various machines, but it seems that for those same testers, BOINC does not recognise the ATI GPU and so cannot run WUs on it. e.g. this host http://www.primegrid.com/show_host_detail.php?hostid=178104 does not show any coprocessors, although the machine has a ATI Radeon 5750. It is running the latest Mac OS 10.7, and has upgraded to the latest BOINC manager (6.13.1). Is this a known limitation or bug in BOINC? As far as I know, there are no other project currently offering OpenCL/Mac apps so I guess it has not been tested before! Any ideas please let me know! Cheers - Iain
4) Message boards : Questions and problems : Cobblestone definition (Message 38643) Posted 23 Jun 2011 by Iain Bethune Post: I'm a little confused after reading the wiki page: http://boinc.berkeley.edu/wiki/Computation_credit The first section has : the Cobblestone (named after Jeff Cobb of SETI@home), is 1/200 day of CPU time on a reference computer that does 1,000 FLOPS i.e. 1 Cobblestone = 432,000 FLOP In the second section : (Remember that a 1 GigaFLOP machine, running full time, produces 200 units of credit in 1 day). i.e. 1 Cobblestone = 432,000,000,000 FLOP I think the first section is incorrect, and should be 1 Cobblestone is 1/200 day on a CPU that does 1,000 MFLOPS (or 1 GFLOP), does that sound right? And if so can someone fix the wiki? Cheers - Iain
5) Message boards : API : Getting a core file from BOINC app (Message 35092) Posted 4 Oct 2010 by Iain Bethune Post: Hi all, I'm trying to debug a problem seen on PrimeGrid with a CUDA app. The problem only seems to affect the Mac (linux and windows from the same code base appear to be OK), and the result is that the app dies with a segfault when it is shutting down (either at the end of the WU, or when being suspended). The problem is, that even when I run the app standalone from the commandline, the BOINC library is catching the SIGSEGV, and reporting a bunch of stuff (executable name, BOINC library version, thread state, images description), when what I really want is just a core file that I can look at with gdb. Is there any way to turn off this behaviour so I can get a core file? Cheers - Iain
6) Message boards : Questions and problems : Mac CUDA Tasks stuck in 'Ready to start' (Message 30778) Posted 3 Feb 2010 by Iain Bethune Post: This could be the problem - is the VRAM requirement something that can be set by the project to stop WU's being downloaded if they can't be run?
7) Message boards : News : BOINC supports GPU computing on Mac OS X (Message 30777) Posted 3 Feb 2010 by Iain Bethune Post: There is also a Mac CUDA app at Primegrid (www.primegrid.com) now too. Cheers - Iain
8) Message boards : Questions and problems : Mac CUDA Tasks stuck in 'Ready to start' (Message 30767) Posted 2 Feb 2010 by Iain Bethune Post: Hi, I'm trying to run the new Mac CUDA app on Primegrid - other users have reported it is working, so I'm guessing it's a client config issue, but I can't figure it out. I've got 'Use GPU always' selected, but the tasks never run. The is the latest alpha version BOINC manager for Mac OS X: 02-Feb-2010 20:18:04 [---] Starting BOINC client version 6.10.29 for x86_64-apple-darwin 02-Feb-2010 20:18:04 [---] log flags: file_xfer, sched_ops, task 02-Feb-2010 20:18:04 [---] Libraries: libcurl/7.19.7 OpenSSL/0.9.7l zlib/1.2.3 c-ares/1.7.0 02-Feb-2010 20:18:04 [---] Data directory: /Library/Application Support/BOINC Data 02-Feb-2010 20:18:04 [---] Processor: 2 GenuineIntel Intel(R) Core(TM)2 Duo CPU T9550 @ 2.66GHz [x86 Family 6 Model 23 Stepping 10] 02-Feb-2010 20:18:04 [---] Processor features: FPU VME DE PSE TSC MSR PAE MCE CX8 APIC SEP MTRR PGE MCA CMOV PAT PSE36 CLFSH DS ACPI MMX FXSR SSE SSE2 SS HTT TM SSE3 MON DSCPL VMX SMX EST TM2 SSSE3 CX16 TPR PDCM SSE4.1 02-Feb-2010 20:18:04 [---] OS: Darwin: 10.2.0 02-Feb-2010 20:18:04 [---] Memory: 4.00 GB physical, 196.20 GB virtual 02-Feb-2010 20:18:04 [---] Disk: 297.77 GB total, 195.96 GB free 02-Feb-2010 20:18:04 [---] Local time is UTC +0 hours 02-Feb-2010 20:18:04 [---] NVIDIA GPU 0: GeForce 9400M (driver version unknown, CUDA version 3000, compute capability 1.1, 254MB, 35 GFLOPS peak) see http://www.pyramid-productions.net/downloads/BOINC.png for a screenshot, showing CPU tasks running, but not GPU tasks. I made a copy of the files in my 'BOINC Data' dir here http://www.pyramid-productions.net/downloads/BOINCData.tar.gz - if that helps to figure out what is set wrong... Any ideas would be appreciated, I'd love to get my GPU crunching! Thanks in advance, - Iain

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.