Problems with OpenCL on OSX 10.10.3 & up, was einsteinbinary_BRP4G_1.39_i686-apple-darwin__BRP4G-opencl-ati-lion
Language
Copyright © 2024 Einstein@Home. All rights reserved.
Copyright © 2024 Einstein@Home. All rights reserved.
Thanks foe asking - ATI
)
Thanks foe asking -
ATI Radeon HD 5770 - two
ATI Radeon R7 250e - one
All 3 were woken fine on 10.9.3
I concur. Though it appeared
)
I concur. Though it appeared at first to be working, a quick check of my validated work units shows that only 1 GPU unit has been successfully completed since the upgrade. The rest appear to be predominantly "aborted by use," and my daily average output remains at approximately 15% of what it was prior to losing the ability to crunch data with GPUs.
I am away from home at this point and cannot provide you with specific GPU info, though I believe 2 of them are AMD Fire Pro D300 (or D500). My main crunching box is a 2014 Mac Pro, so the hardware is definitely not obsolete.
BOINC 7.6.10 not working well
)
BOINC 7.6.10 not working well on a Mac Pro with an ATI 5770 and El Capitan GM seed 1; nothing but computation errors on the GPU work units.
Computation errors show up after about 30 seconds of the units being downloaded. Jumps to 100% then errors.
CUDA and nVidia cards work just fine though.
RE: Thanks foe asking
)
Sorry about the misspellings! Hopefully iOS 9 will save me in the future. I meant to say "Thanks for asking" and " All 3 were working fine on 10.9.3".
My cuda machines are down so my Radeon machine is my one and only, and I don't have a way to downgrade back to 10.9. So I am not able to help Einstein until this issue is resolved. This issue gives the impression that Mac's are second class citizens and is very discouraging.
RE: This issue gives the
)
Mac defined as second class by whom? Us? No. The problem are the upstream vendors who simply treat OpenCL as second class citizen...
Oliver
Einstein@Home Project
Guess so, since BOINC says
)
Guess so, since BOINC says OpenCL is still running at version 1.2 even in El Capitan. ATI probably feels it doesn't need to update OpenCL for Apple's older GPUs, it's like the 5770 which was a staple for all of its last generation of Mac Pros is dead in the water.
OS X El Capitan released
)
OS X El Capitan released today. Had to reinstall BOINC, v7.6.11, getting endless AMD/ATI computation errors. Guess it's just a matter of patience..
Isn´t it kind of usual to
)
Isn´t it kind of usual to have to reinstall? After a new release install, it´s not the first time that I see a message for it to be needed due to permissions not set OK.
However, I don´t keep BOINC install files on my HD forever, usually delete them right after that. So I just downloaded them from Berkeley. Amazingly, it appeared to be a newer one (7.6.12), they come in rapid succession at the moment! I don´t know what it is about, as it´s undocumented since 7.6.9; alas, it´s dark as night there since it´s not under Dr. Anderson´s responsibility anymore... Maybe related to your errors, maybe not, just thought I´d mention it.
Another thing is that FGRP4 executables changed to version 1.15 after the short outage. Again, maybe related, maybe not.
RE: Another thing is that
)
Nope, that was just a scientific bug fix which also required to let the task pool to run dry.
Oliver
Einstein@Home Project
Okay, that's really weird..
)
Okay, that's really weird.. Parkes PMPS XT / BRP6 Beta v1.52 tasks are working just fine.. it's the BRP4G beta / Arecibo v1.39 tasks that immediately go to 100% and computation error.
So I deactivated Arecibo in Einstein prefs and aborted all the Arecibo tasks in my queue and it's all good now.
Sorry for the unjustified kvetching.
Edit: waait a sec.. Parkes updated to v.1.58 after a few refreshes.. well it's still going so all righty then. So Parkes PMPS XT / BRP6 Beta is fine, it's just Arecibo that's not working, although I bet Arecibo won't ever be updated as far as I understand because there's no more raw data coming from Arecibo?