I reported both the MacBook pro problems to the Boinc Alpha and Dev lists earlier (I only ment to do it to the Boinc Alpha list), hoping Charlie will come and have a look, no replies yet.
> So at least this is boinc related and not specific to Einstein.
It doesn't have to be and I don't really think it is. It really appears to be an issue of Apple's OpenCL framework in conjunction with GPU switching. If you can afford it diskspace-wise, I recommend you install OS X 10.9 Mavericks in parallel to your current OS. This way you can give it a spin without affecting your current installation. See if it fixes the problem.
Update: we tried to reproduce the issue on a MBP running 10.7.5. So far we only managed to switch from internal GPU to discrete GPU while BOINC was running our app - that worked fine. It remains to be seen what happens if we switch from discrete to internal while running our OpenCL app, but we weren't yet able to reproduce that state change...
Update: we tried to reproduce the issue on a MBP running 10.7.5. So far we only managed to switch from internal GPU to discrete GPU while BOINC was running our app - that worked fine.
That sounds like you have replicated my situation, meaning the problem is somehow related to my GPU. Could you post the start of the boinc log for the record?
Update: we tried to reproduce the issue on a MBP running 10.7.5. So far we only managed to switch from internal GPU to discrete GPU while BOINC was running our app - that worked fine.
That sounds like you have replicated my situation, meaning the problem is somehow related to my GPU. Could you post the start of the boinc log for the record?
Not quite, we set the GPU to internal (fixed), started our app in BOINC and only then switched to the discrete GPU. Thus we didn't rely on automatic switching. We'll look into it and post the updates as soon as we have them (incl. the system specs as seen by BOINC, hint: we also see the 10.7.4 on a 10.7.5 system).
Did a re-install of boinc to fix the permission issues, then allowed
work from einstein and "Use GPU always" with the integrated card active.
The top of the boinc startup now contains:
06-Feb-2014 08:46:33 [---] Starting BOINC client version 7.2.33 for x86_64-apple-darwin
06-Feb-2014 08:46:33 [---] log flags: file_xfer, sched_ops, task
06-Feb-2014 08:46:33 [---] Libraries: libcurl/7.26.0 OpenSSL/1.0.1e zlib/1.2.5 c-ares/1.9.1
06-Feb-2014 08:46:33 [---] Data directory: /Library/Application Support/BOINC Data
06-Feb-2014 08:46:33 [---] OpenCL: AMD/ATI GPU 0: ATI Radeon HD 6750M (driver version 1.2 (Sep 19 2013 22:22:06), device version OpenCL 1.2, 512MB, 512MB available, 72 GFLOPS peak)
06-Feb-2014 08:46:33 [---] OpenCL CPU: Intel(R) Core(TM) i7-2675QM CPU @ 2.20GHz (OpenCL driver vendor: Apple, driver version 1.1, device version OpenCL 1.2)
So it seems I have some new versions of openCL. The first task is now
running, but gfxCardStatus reports the integrated graphics chip is
active! The stderr is:
more /Library/Application\ Support/BOINC\ Data/slots/5/stderr.txt
[08:49:49][63509][INFO ] Application startup - thank you for supporting Einstein@Home!
[08:49:49][63509][INFO ] Starting data processing...
[08:49:49][63509][INFO ] Using OpenCL platform provided by: Apple
[08:49:49][63509][INFO ] Using OpenCL device "ATI Radeon HD 6750M" by: AMD
[08:49:51][63509][INFO ] Checkpoint file unavailable: status.cpt (No such file or directory).
------> Starting from scratch...
[08:49:51][63509][INFO ] Header contents:
------> Original WAPP file: ./PA0062_005A1_DM2244.00
------> Sample time in microseconds: 1000
------> Observation time in seconds: 2097.152
------> Time stamp (MJD): 54051.706071142245
------> Number of samples/record: 0
------> Center freq in MHz: 1231.5
------> Channel band in MHz: 3
------> Number of channels/record: 96
------> Nifs: 1
------> RA (J2000): 71245.5339012
------> DEC (J2000): -55940.8699989
------> Galactic l: 0
------> Galactic b: 0
------> Name: G4403510
------> Lagformat: 0
------> Sum: 1
------> Level: 3
------> AZ at start: 0
------> ZA at start: 0
------> AST at start: 0
------> LST at start: 0
------> Project ID: --
------> Observers: --
------> File size (bytes): 0
------> Data size (bytes): 0
------> Number of samples: 2097152
------> Trial dispersion measure: 2244 cm^-3 pc
------> Scale factor: 1.76471
[08:49:52][63509][INFO ] Seed for random number generator is 1081990622.
[08:49:56][63509][ERROR] Error in OpenCL context: OpenCL Build Warning : Compiler build log:
:199:23: warning: double precision constant requires cl_khr_fp64, casting to single precision
ang = 2.0f * M_PI * dir * i * (startRow + j) / N; \
^
:4:14: note: expanded from macro 'M_PI'
#define M_PI 0x1.921fb54442d18p+1
^
:431:1: warning: unused variable 'c'
Activity monitor is reporting that the task takes around ~10% cpu time.
I have tried to force the AMD discrete card active while the task is
running, but I see no change in behavior. In particular I would expect
the screen to lag while I am using it, but everything seems smooth. Want
to enlighten me in this new behavior? Does boinc utilize the non-active
GPU?
I reported both the MacBook
)
I reported both the MacBook pro problems to the Boinc Alpha and Dev lists earlier (I only ment to do it to the Boinc Alpha list), hoping Charlie will come and have a look, no replies yet.
Claggy
Hi Tullus, > So at least
)
Hi Tullus,
> So at least this is boinc related and not specific to Einstein.
It doesn't have to be and I don't really think it is. It really appears to be an issue of Apple's OpenCL framework in conjunction with GPU switching. If you can afford it diskspace-wise, I recommend you install OS X 10.9 Mavericks in parallel to your current OS. This way you can give it a spin without affecting your current installation. See if it fixes the problem.
Update: we tried to reproduce the issue on a MBP running 10.7.5. So far we only managed to switch from internal GPU to discrete GPU while BOINC was running our app - that worked fine. It remains to be seen what happens if we switch from discrete to internal while running our OpenCL app, but we weren't yet able to reproduce that state change...
Best,
Oliver
Einstein@Home Project
RE: Update: we tried to
)
That sounds like you have replicated my situation, meaning the problem is somehow related to my GPU. Could you post the start of the boinc log for the record?
I will let you know if I take the jump to mavericks (issues like this: http://apple.stackexchange.com/questions/106796/how-do-i-sync-my-iphone-3g-to-os-x-10-9 is highly annoying...). I already have a bootable carbon copy, so I guess jumping back and forth should be doable.
RE: RE: Update: we tried
)
Not quite, we set the GPU to internal (fixed), started our app in BOINC and only then switched to the discrete GPU. Thus we didn't rely on automatic switching. We'll look into it and post the updates as soon as we have them (incl. the system specs as seen by BOINC, hint: we also see the 10.7.4 on a 10.7.5 system).
Stay tuned,
Oliver
Einstein@Home Project
I have finally made the move
)
I have finally made the move over to mavericks!
Did a re-install of boinc to fix the permission issues, then allowed
work from einstein and "Use GPU always" with the integrated card active.
The top of the boinc startup now contains:
So it seems I have some new versions of openCL. The first task is now
running, but gfxCardStatus reports the integrated graphics chip is
active! The stderr is:
more /Library/Application\ Support/BOINC\ Data/slots/5/stderr.txt [08:49:49][63509][INFO ] Application startup - thank you for supporting Einstein@Home! [08:49:49][63509][INFO ] Starting data processing... [08:49:49][63509][INFO ] Using OpenCL platform provided by: Apple [08:49:49][63509][INFO ] Using OpenCL device "ATI Radeon HD 6750M" by: AMD [08:49:51][63509][INFO ] Checkpoint file unavailable: status.cpt (No such file or directory). ------> Starting from scratch... [08:49:51][63509][INFO ] Header contents: ------> Original WAPP file: ./PA0062_005A1_DM2244.00 ------> Sample time in microseconds: 1000 ------> Observation time in seconds: 2097.152 ------> Time stamp (MJD): 54051.706071142245 ------> Number of samples/record: 0 ------> Center freq in MHz: 1231.5 ------> Channel band in MHz: 3 ------> Number of channels/record: 96 ------> Nifs: 1 ------> RA (J2000): 71245.5339012 ------> DEC (J2000): -55940.8699989 ------> Galactic l: 0 ------> Galactic b: 0 ------> Name: G4403510 ------> Lagformat: 0 ------> Sum: 1 ------> Level: 3 ------> AZ at start: 0 ------> ZA at start: 0 ------> AST at start: 0 ------> LST at start: 0 ------> Project ID: -- ------> Observers: -- ------> File size (bytes): 0 ------> Data size (bytes): 0 ------> Number of samples: 2097152 ------> Trial dispersion measure: 2244 cm^-3 pc ------> Scale factor: 1.76471 [08:49:52][63509][INFO ] Seed for random number generator is 1081990622. [08:49:56][63509][ERROR] Error in OpenCL context: OpenCL Build Warning : Compiler build log: :199:23: warning: double precision constant requires cl_khr_fp64, casting to single precision ang = 2.0f * M_PI * dir * i * (startRow + j) / N; \ ^ :4:14: note: expanded from macro 'M_PI' #define M_PI 0x1.921fb54442d18p+1 ^ :431:1: warning: unused variable 'c'[....] sniped [...]
[08:49:57][63509][INFO ] Derived global search parameters:
------> f_A probability = 0.04
------> single bin prob(P_noise > P_thr) = 1.2977e-08
------> thr1 = 18.1601
------> thr2 = 21.263
------> thr4 = 26.2923
------> thr8 = 34.674
------> thr16 = 48.9881
[08:59:53][63509][INFO ] Checkpoint committed!
[09:09:57][63509][INFO ] Checkpoint committed!
Activity monitor is reporting that the task takes around ~10% cpu time.
I have tried to force the AMD discrete card active while the task is
running, but I see no change in behavior. In particular I would expect
the screen to lag while I am using it, but everything seems smooth. Want
to enlighten me in this new behavior? Does boinc utilize the non-active
GPU?