Problems with OpenCL on OSX 10.10.3 & up, was einsteinbinary_BRP4G_1.39_i686-apple-darwin__BRP4G-opencl-ati-lion

Cap
Cap
Joined: 21 Feb 05
Posts: 5
Credit: 524128434
RAC: 12
Topic 198054

The app continually crashes on yosemite, 10.10.3

Moldr
Moldr
Joined: 3 Apr 15
Posts: 11
Credit: 145119
RAC: 0

Problems with OpenCL on OSX 10.10.3 & up, was einsteinbinary_BRP

Quote:
The app continually crashes on yosemite, 10.10.3

I don't know if my situation is better or worse than yours. On my machine the app is taking 3-4 times as long as normal to complete. Right now remaining time (estimated) is 01:15:16 with .015%, that's right 0.015%, left having slowed to a crawl. It was rolling along okay and progressively slowed down. I just upgraded to 10.10.3 from 10.9.5 last night and I'm wondering if that had anything to do with it.

Edit: I also forgot to mention that the app is now using 50% of one core and, apparently, causing kernel task to use 50% as well. I hadn't noticed that behavior prior to upgrading to 10.10.3. I can't be certain about kernel task but I'm sure the app itself was not using 50% constantly when I was on 10.9.5.

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171438
RAC: 28

Hi guys, Did you already

Hi guys,

Did you already install the 10.10.3 supplemental update which fixes a graphics driver issue? It might be related. I don't see any issue with the app on my 10.10.3 iMac having the update installed. A new BRP4G task's expected to be finished in about 80 minutes (Radeon 6970M, BOINC 7.4.36). But yes, I can confirm the rather high 50% CPU load for that task. That's something we need to look into...

Cheers,
Oliver

Einstein@Home Project

Moldr
Moldr
Joined: 3 Apr 15
Posts: 11
Credit: 145119
RAC: 0

Oliver, Thanks for the reply.

Oliver,
Thanks for the reply. i have installed the driver update and restarted both the OS and BOINC. I then tried to run a BRP4G task and had a 50% load on CPU and a 50% load on kernel task process while it was running. I let it run for over 16 hours and it never completed, prior to updating to Yosemite my complete times on BRP4G tasks usually ran ~5 hours. It reached 99.999% completed and ran for over an hour after reaching that point so I aborted. In addition, it appears that no checkpointing was occurring during the task. I checked under the properties command, and CPU time at last checkpoint was blank, and if BOINC manager was exited and restarted the task would restart from the beginning regardless of the apparent completion percentage prior to exiting.

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171438
RAC: 28

Which BOINC version do you

Which BOINC version do you use?

Einstein@Home Project

Moldr
Moldr
Joined: 3 Apr 15
Posts: 11
Credit: 145119
RAC: 0

Version 7.4.42

Version 7.4.42

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171438
RAC: 28

Ok, we'll look into this

Ok, we'll look into this after the weekend...

Einstein@Home Project

Bill & Patsy
Bill & Patsy
Joined: 8 Sep 07
Posts: 17
Credit: 5242914
RAC: 0

Yes. Both recent application

Yes. Both recent application upgrades have broken my computer too (Computer ID: 8479367). Specifically, (Arecibo, GPU) v1.39 (BRP4G-opencl-ati-lion) and (Parkes PMPS XT) v1.52 (BRP6-opencl-ati-lion) are both broken on my machine. v1.39 has symptoms like Cap and Moldr describe. v1.52 crashes instantly.

Bummer.

And yes, I'm running the very latest updates on everything.

--Bill

Moldr
Moldr
Joined: 3 Apr 15
Posts: 11
Credit: 145119
RAC: 0

Just to add a little more

Just to add a little more information. I re-enabled GPU tasks and (BRP6-opencl-ati-lion) because I had one complete 14 April, it was v1.41. Using v1.52 I can also corroborate Cap and Bill and Patsy's results, v1.52 errors out within seconds of starting.

Richard Haselgrove
Richard Haselgrove
Joined: 10 Dec 05
Posts: 2143
Credit: 2959042844
RAC: 710159

RE: Ok, we'll look into

Quote:
Ok, we'll look into this after the weekend...


v7.4.42 displays an estimate of progress immediately, based on the initial estimate of total runtime, and increases that estimate asymptotically towards 100%. I wrote this up as "pseudo-progress" for FGRP4.

Pseudo-progress is replaced by real reported progress in the display at the first checkpoint. So Moldr's observations are consistent with the application never starting any real computation at all. It may be helpful to inspect the contents of the slot directory, starting with stderr.txt

Oliver Behnke
Oliver Behnke
Moderator
Administrator
Joined: 4 Sep 07
Posts: 984
Credit: 25171438
RAC: 28

RE: I don't see any issue

Quote:
I don't see any issue with the app on my 10.10.3 iMac having the update installed. A new BRP4G task's expected to be finished in about 80 minutes (Radeon 6970M, BOINC 7.4.36).

I need to revert that statement, that task went down the drain (more than a day of runtime, high CPU load, no checkpointing, eventual segfault) on my box as well! Good, that gives us something to work with :-)

Stay tuned...

Einstein@Home Project

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.