MacBook Pro Retina / CUDA doesn't work
6 Oct 2012 15:35:13 UTC
Topic 196558
(moderation:
HI,
my MacBook is equipped with two graphics: Intel HD4000 and Nvidia GT 650M. In the settings in MacOS I can select that the Nvidia has to be used all the time instead of switching between the two cards.
In the BOINC messages the Nvidia isn't detected. That's what it shows:
Sa 6 Okt 17:15:27 2012 | | Starting BOINC client version 7.0.31 for x86_64-apple-darwin
Sa 6 Okt 17:15:27 2012 | | log flags: file_xfer, sched_ops, task
Sa 6 Okt 17:15:27 2012 | | Libraries: libcurl/7.26.0 OpenSSL/0.9.7l zlib/1.2.5 c-ares/1.9.1
Sa 6 Okt 17:15:27 2012 | | Data directory: /Library/Application Support/BOINC Data
Sa 6 Okt 17:15:27 2012 | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-3615QM CPU @ 2.30GHz [x86 Family 6 Model 58 Stepping 9]
Sa 6 Okt 17:15:27 2012 | | 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 PBE SSE3 PCLMULQDQ DTES64 MON DSCPL VMX EST TM2 SSSE3 CX16 TPR PDCM SSE4.1 SSE4.2 xAPIC POPCNT AES PCID XSAVE OSXSAVE TSCTMR AVX1.0 RDRAND F16C
Sa 6 Okt 17:15:27 2012 | | OS: Mac OS X 10.8.2 (Darwin 12.2.0)
Sa 6 Okt 17:15:27 2012 | | Memory: 16.00 GB physical, 98.33 GB virtual
Sa 6 Okt 17:15:27 2012 | | Disk: 232.96 GB total, 98.09 GB free
Sa 6 Okt 17:15:27 2012 | | Local time is UTC +2 hours
Sa 6 Okt 17:15:27 2012 | | OpenCL: NVIDIA GPU 0: GeForce GT 650M (driver version CLH 1.0, device version OpenCL 1.1, 1024MB, 1024MB available)
Sa 6 Okt 17:15:27 2012 | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID not assigned yet; resource share 100
What can I do to make my Nvidia compute? Thanks for Help!
Proud member of SETI.Germany
MacBook Pro Retina / CUDA doesn't work
)
I see:
However, the only OpenCL capable applications here at Einstein are at this time for AMD GPUs only. See http://einstein.phys.uwm.edu/apps.php
So that makes you wonder why your GPU's CUDA capability isn't being detected and that can lie in the fact that you forgot to install the Nvidia manufacturer drivers, see http://developer.nvidia.com/cuda/cuda-downloads for that. The drivers built into OS X won't contain things like CUDA.
There were CUDA-drivers
)
There were CUDA-drivers installed, but they were in version 5.0.17. I updated them to 5.0.36 and BOINC detects something:
After hat Einstein starts some downloads brings that:
After three times I stopped it. What is next to do?
Proud member of SETI.Germany
Seeing the actual
)
Seeing the actual error:
process exited with code 234 (0xea, -22)
dyld: DYLD_ environment variables being ignored because main executable (/Library/Application Support/BOINC Data/slots/8/../../switcher/switcher) is setuid or setgid
[18:52:32][3251][INFO ] Application startup - thank you for supporting Einstein@Home!
[18:52:32][3251][INFO ] Starting data processing...
[18:52:32][3251][ERROR] Failed to enable CUDA thread yielding for device #0 (error: 2)! Sorry, will try to occupy one CPU core...
[18:52:32][3251][ERROR] Couldn't acquire CUDA context of device #0 (error: 2)!
[18:52:32][3251][ERROR] Demodulation failed (error: 1002)!
18:52:32 (3251): called boinc_finish
]]>
It looks like the science application cannot find a CUDA capable GPU (it checks separately). This could be due to wrong installation of drivers, or that you really do need the whole thing of Tools, driver + SDK as it shows in the link I gave you earlier. I'm not a Mac person, so other than that...
Edit: read also the last lines of http://boinc.berkeley.edu/wiki/GPU_computing#Things_to_be_aware_of; How did you permanently set the GPU to the Nvidia device? As it says there, or differently?
"This release of the CUDA
)
"This release of the CUDA Toolkit enables development using GPUs using the Kepler architecture..."
My GT 650M is a kepler-based device, so I will try it this way. I can select in the system settings that I want the Nvidia to be permanently used by checking a box. By unchecking the box the Mac decides itself which GPU will be used.
EDIT: BOINC starts computing and it's running since a few minutes. Tomorrow I'll tell you whether it has worked during the night.
Proud member of SETI.Germany
It's running. But now it
)
It's running. But now it stops and says:
The GPU-memory is 1GB, maybe it's full, I don't know how to check this. But my normal memory, 16GB, is about half empty. Is there something I can do?
Proud member of SETI.Germany
It's possible that the past
)
It's possible that the past crashes left artifacts in memory. You can only clean that by rebooting, so try that first.
Seems to work. During the
)
Seems to work. During the night I completed 15 tasks. Thank you for the support!
Proud member of SETI.Germany