I don't mind any debugging I can do, if you don't mind sending me detailed instructions. I'm afraid I'm pretty new at Macs.
I did download the CUDA software in your last post, and installed it. This time, my log says:
Mon Nov 18 14:35:24 2013 | Einstein@Home | Sending scheduler request: To fetch work.
Mon Nov 18 14:35:24 2013 | Einstein@Home | Requesting new tasks for CPU and NVIDIA and intel_gpu
Mon Nov 18 14:35:30 2013 | Einstein@Home | Scheduler request completed: got 1 new tasks
Mon Nov 18 14:36:31 2013 | Einstein@Home | Sending scheduler request: To fetch work.
Mon Nov 18 14:36:31 2013 | Einstein@Home | Requesting new tasks for CPU and NVIDIA and intel_gpu
Mon Nov 18 14:36:35 2013 | Einstein@Home | Scheduler request completed: got 1 new tasks
Mon Nov 18 14:37:35 2013 | Einstein@Home | Sending scheduler request: To fetch work.
Mon Nov 18 14:37:35 2013 | Einstein@Home | Requesting new tasks for NVIDIA and intel_gpu
Mon Nov 18 14:37:36 2013 | Einstein@Home | Scheduler request completed: got 0 new tasks
Mon Nov 18 14:37:36 2013 | Einstein@Home | No work sent
Mon Nov 18 14:37:36 2013 | Einstein@Home | see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/9284/9284410
Mon Nov 18 14:37:36 2013 | Einstein@Home | Binary Radio Pulsar Search (Arecibo) is not available for your type of computer.
I downloaded the driver you suggested. I still don't get any tasks. It looks like it doesn't see the CUDA driver. But I did install and reboot as you suggested. The scheduler log still shows:
Mon Nov 18 08:21:38 2013 | | No config file found - using defaults
Mon Nov 18 08:21:38 2013 | | Starting BOINC client version 7.0.65 for x86_64-apple-darwin
Mon Nov 18 08:21:38 2013 | | log flags: file_xfer, sched_ops, task
Mon Nov 18 08:21:38 2013 | | Libraries: libcurl/7.26.0 OpenSSL/1.0.1c zlib/1.2.5 c-ares/1.9.1
Mon Nov 18 08:21:38 2013 | | Data directory: /Library/Application Support/BOINC Data
Mon Nov 18 08:21:38 2013 | | Processor: 8 GenuineIntel Intel(R) Core(TM) i7-4850HQ CPU @ 2.30GHz [x86 Family 6 Model 70 Stepping 1]
Mon Nov 18 08:21:38 2013 | | 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 SMX EST TM2 SSSE3 FMA CX16 TPR PDCM SSE4.1 SSE4.2 x2APIC MOVBE POPCNT AES PCID XSAVE OSXSAVE SEGLIM64 TSCTMR AVX1.0 RDRAND F16C
Mon Nov 18 08:21:38 2013 | | OS: Mac OS X 10.9.0 (Darwin 13.0.2)
Mon Nov 18 08:21:38 2013 | | Memory: 16.00 GB physical, 378.27 GB virtual
Mon Nov 18 08:21:38 2013 | | Disk: 465.12 GB total, 378.02 GB free
Mon Nov 18 08:21:38 2013 | | Local time is UTC -5 hours
Mon Nov 18 08:21:38 2013 | | BOINC does not support CUDA on this computer.
Mon Nov 18 08:21:38 2013 | | OpenCL: NVIDIA GPU 0: GeForce GT 750M (driver version 8.20.15 310.40.15f03, device version OpenCL 1.2, 2048MB, 2048MB available, 30 GFLOPS peak)
Mon Nov 18 08:21:38 2013 | | OpenCL: Intel GPU 0: Iris Pro (driver version 1.2(Sep 29 2013 19:45:54), device version OpenCL 1.2, 1024MB, 1024MB available, 2688 GFLOPS peak)
You now are showing a message that Cuda support isn't available, see the earlier post about Cuda support being disabled on dual-GPU MacBook Pros. (you're got a Nvidia and an Intel GPU)
What does Boinc 7.2.28 show? Charlie may have re-enabled Cuda support on that version, it does have the following changeset:
Quote:
client: optionally detect GPUs via a child process, for dual_GPU laptops.
Some dual-GPU laptops (e.g., Macbook Pro) don't power down the more powerful GPU until all applications which used them exit. To save battery life, the client launches a second instance of the client as a child process to detect and get info about the GPUs. The child process writes the info to a temp file which our main client then reads. This option is enabled at compile time by defining USE_CHILD_PROCESS_TO_DETECT_GPUS as non-zero in gpu_detect.cpp
Thanks Claggy, I didn't pay enough attention to the BOINC version Richard was using.
For the sake of completeness: I tested 5.5.28 on a OSX 10.9 machine with a Quadro 4000 (> sm_1x GPU) and BOINC 7.2.28 - it's being recognised without a problem.
To conclude: to get CUDA 5.5 working on MacBook Pros with OSX 10.9 Mavericks you need to:
* Install the latest CUDA driver from NVIDIA (CUDA not supported by stock driver shipped by Apple)
* Install the latest BOINC client software
* For the time being: use a > sm_1x GPU (Fermi and newer) until NVIDIA releases a driver update to support older GPUs again
But you're erroring Cuda Wu's with the following now:
Quote:
[05:05:45][11814][INFO ] CUDA global memory status (GPU setup complete):
------> Used in total: 340 MB (1709 MB free / 2049 MB total) -> Used by this application (assuming a single GPU task): 122 MB
[05:05:50][11814][INFO ] Checkpoint committed! [06:03:01][11814][ERROR] Error during CUDA device->host time series length transfer (error: 700)
[06:03:01][11814][ERROR] Demodulation failed (error: 1008)!
06:03:01 (11814): called boinc_finish
Error 700 often means that the CUDA device context was invalidated for some reason. Richard, did you try to switch the GPU manually or can you think of a situation where OSX might have tried to do that?
Richard, you might want to try this tool. It allows you to monitor which GPU is being used. It also allows you to set the active GPU manually, disabling automatic switching. It would be interesting to know whether the errors you get still appear when you manually switch to the dedicated GPU before you start BOINC (any GPU app in BOINC). This way we should get an indication whether the errors are caused by GPU switching or not...
I don't mind any debugging I
)
I don't mind any debugging I can do, if you don't mind sending me detailed instructions. I'm afraid I'm pretty new at Macs.
I did download the CUDA software in your last post, and installed it. This time, my log says:
Mon Nov 18 14:35:24 2013 | Einstein@Home | Sending scheduler request: To fetch work.
Mon Nov 18 14:35:24 2013 | Einstein@Home | Requesting new tasks for CPU and NVIDIA and intel_gpu
Mon Nov 18 14:35:30 2013 | Einstein@Home | Scheduler request completed: got 1 new tasks
Mon Nov 18 14:36:31 2013 | Einstein@Home | Sending scheduler request: To fetch work.
Mon Nov 18 14:36:31 2013 | Einstein@Home | Requesting new tasks for CPU and NVIDIA and intel_gpu
Mon Nov 18 14:36:35 2013 | Einstein@Home | Scheduler request completed: got 1 new tasks
Mon Nov 18 14:37:35 2013 | Einstein@Home | Sending scheduler request: To fetch work.
Mon Nov 18 14:37:35 2013 | Einstein@Home | Requesting new tasks for NVIDIA and intel_gpu
Mon Nov 18 14:37:36 2013 | Einstein@Home | Scheduler request completed: got 0 new tasks
Mon Nov 18 14:37:36 2013 | Einstein@Home | No work sent
Mon Nov 18 14:37:36 2013 | Einstein@Home | see scheduler log messages on http://einstein.phys.uwm.edu//host_sched_logs/9284/9284410
Mon Nov 18 14:37:36 2013 | Einstein@Home | Binary Radio Pulsar Search (Arecibo) is not available for your type of computer.
RE: Thanks Oliver. I
)
You now are showing a message that Cuda support isn't available, see the earlier post about Cuda support being disabled on dual-GPU MacBook Pros. (you're got a Nvidia and an Intel GPU)
What does Boinc 7.2.28 show? Charlie may have re-enabled Cuda support on that version, it does have the following changeset:
Claggy
Bingo! That was it. My
)
Bingo!
That was it. My macbook is now happily crunching away.
Thanks to all who replied.
Thanks Claggy, I didn't pay
)
Thanks Claggy, I didn't pay enough attention to the BOINC version Richard was using.
For the sake of completeness: I tested 5.5.28 on a OSX 10.9 machine with a Quadro 4000 (> sm_1x GPU) and BOINC 7.2.28 - it's being recognised without a problem.
To conclude: to get CUDA 5.5 working on MacBook Pros with OSX 10.9 Mavericks you need to:
* Install the latest BOINC client software
* For the time being: use a > sm_1x GPU (Fermi and newer) until NVIDIA releases a driver update to support older GPUs again
Cheers,
Oliver
Einstein@Home Project
But you're erroring Cuda Wu's
)
But you're erroring Cuda Wu's with the following now:
Claggy
FYI, "you" in Claggy's port
)
FYI, "you" in Claggy's port most likely means this host: http://einsteinathome.org/host/9284410/tasks.
Error 700 often means that the CUDA device context was invalidated for some reason. Richard, did you try to switch the GPU manually or can you think of a situation where OSX might have tried to do that?
Oliver
Einstein@Home Project
No, absolutely not. In fact,
)
No, absolutely not. In fact, I don't think I could if I wanted to.
I see that one cuda wu just went to a ready to report status. No sign of an error here, but let's see what this shows, when it reports.
RE: I see that one cuda wu
)
These are your errored tasks:
Error tasks for computer 9284410
Edit: I've reported the three Mac GPU error threads to the Boinc Alpha list in hope that the Mac developer comes along and has a look.
Claggy
Richard, you might want to
)
Richard, you might want to try this tool. It allows you to monitor which GPU is being used. It also allows you to set the active GPU manually, disabling automatic switching. It would be interesting to know whether the errors you get still appear when you manually switch to the dedicated GPU before you start BOINC (any GPU app in BOINC). This way we should get an indication whether the errors are caused by GPU switching or not...
Best,
Oliver
Einstein@Home Project
Thanks. I installed the app
)
Thanks. I installed the app and have my GPU set to just the nvidia now.
Let's see what happens next time I report a cuds wu.