BRP5 on iGPU
3 Aug 2013 13:31:30 UTC
Topic 197111
(moderation:
Seems E@H have decided that BRP5 can now be crunched on iGPU.
I have a CPU with iGPU and an external ATI card.
It used to be that BRP5 ran on the ATI card, BRP4 on the iGPU and the rest on the CPU.
Now the iGPU has started crunching BRP5 units as well as BRP4 units. 23% after 5 hours (with 3 concurrent tasks). Compared to 30 minutes for 3 BRP4 tasks. My ATI card needs 2:40 hours to do 3 BRP5 tasks.
I would like to use my powerful ATI for the big BRP5 tasks only, and the weak iGPU for BRP4 only. Any way to accomplish this?
BRP5 on iGPU
)
Another more serious issue with BRP5 on iGPU:
After having completed the first iGPU BRP5 units boinc raised the completion times on all tasks massively. Since the BRP5 unit all of a sudden took much longer boinc assumed all resources on my computer has become much slower I guess.
This had the unfortunate effect that my CPU units (FGRP2 and S6CasA) now is set with 72 hours completion time and has gotten "high priority". This means that all CPU cores are now crunching CPU jobs and my GPU jobs are on hold since no CPU resources are available to them. Effectively turning off my GPUs
Maybe this will correct it self after a time (weeks?) but it seems to me that boinc can't handle having the same machine completing the same job (BRP5) in 2 massively different times (iGPU and real GPU). Making me think that E@H should revert whatever change you did to make iGPU crunch BRP5.
Only thing I can do is to disable my iGPU but that would be a shame since it crunched the BRP4 units nicely.
I'm seeing the same thing and
)
I'm seeing the same thing and would also like to ask that the project revert this change.
For now I'm planning on running down my cache of BRP5 tasks assigned to my HD 4000 and then see if it's possible to set up a cc_config.xml with the option to not receive any more BRP5 task on my iGPU.
I'm thinking something like this might work:
[pre]
http://einstein.phys.uwm.edu/
0
intel_gpu
einsteinbinary_BRP5
[/pre]
It will take a few more days before I'm ready to test this.
RE: This means that all CPU
)
Have you tried to go into your Boinc preferences (Tools > Computing Preferences > Processor usage) and reduced the percentage of processors used?
Disclaimer: I'm not 100% certain this will solve the problem you're having, but it's worth a try?
AM
RE: For now I'm planning on
)
Sounds like a nice way forward. I'm a pedestrian when it comes to those config files so if it works please post an example for my copy/paste pleasure :)
First thing I tried :)
That will only change the number of cores running CPU jobs. Since the CPU jobs have "high priority" they get all available CPU resources.
On the bright side of things it seems boinc have already lowered the estimated time on CPU jobs to around 30 hours which made them run on normal priority so my GPUs are back in business.
RE: On the bright side of
)
And now we're back up to 100 hour estimates on the CPU jobs. Meaning high priority and no GPU work being done.
I guess this will keep happening every time the iGPU finishes a BRP5 job. Not acceptable.
While we wait for a
)
While we wait for a resolution you might want to lower your cache settings to avoid high priority. I'm running with 0.3 + 0.2 days and none of my tasks have run at high priority.
BRP5 on iGPU has been
)
BRP5 on iGPU has been deprecated (disabled).
BM
BM
Thank you Bernd!
)
Thank you Bernd!
Thanks. I aborted the
)
Thanks. I aborted the remaining iGPU BRP5 tasks and now things are running smoothly.
What a mess, who would have
)
What a mess, who would have thought that! My estimated CPU runtimes also increased, but running a ~1 day cache setting this luckily didn't trigger panic mode of my BOINC.
I just which BOINC would set, treat and schedule things per ressource rather than per project / app. It's imposible to have one result duration correction factor apply to all co-/processors at once..
MrS
Scanning for our furry friends since Jan 2002