Gamma-ray pulsar search #3
2 Feb 2014 1:14:20 UTC
Topic 197368
(moderation:
I just processed 1 of these on my GPU on my W8 box.
I have it set to leave 1 core free to feed the GPU and run CPU tasks on the other 3 cores. In all I was running 6 tasks at once, all fine. When this little devil arrived it took one of the 3 cores I reserve for CPU processing and of course I had not set the GPU to run 3 at a time since I did not know there was a GPU app for this data. so I ended up running only 3 apps. The run time was only slightly shorter than my CPU and the credit was the same.
How can I restrict this data to the CPU and run BRPs on the GPU?
Gamma-ray pulsar search #3
)
Set what applications you want to run at http://einstein.phys.uwm.edu/prefs.php?subset=project
I do not see the option of
)
I do not see the option of opting out of GPU processing for Gamma-ray pulsar search #3 and only CPU processing. As I see it, it is all or nothing.
According to this message
)
According to this message http://einsteinathome.org/node/197343&nowrap=true#128786 you should not get gpu wu's .
Or did I miss something?
You could set the GPU
)
You could set the GPU utilization settings for FGRP tasks to 0 in settings and tick the option of preference for cpu work also in settings. As far as i know the FGRP GPU tasks were suspended because of an issue of the OpenCL version on which they are working.
RE: You could set the GPU
)
According to this message http://einsteinathome.org/node/197343&nowrap=true#128671 this is a known bug and should be fixed in the next version.
Thanks, I'll try that.
)
Thanks, I'll try that.
RE: I just processed 1 of
)
When the FGRP3 run started, there were a couple of small releases of tasks (as a sort of trial, I guess) before the flood gates were opened. In those small releases, there were both CPU and GPU tasks and you obviously got two of the latter as they are showing in your tasks list. As already pointed out, the GPU app has been withdrawn for the moment so you won't get any more until there is a further announcement.
Unfortunately, I don't think you can. I want to do exactly the same thing and for the moment, the withdrawal of the GPU app has saved things temporarily. I asked for separate check boxes for the two 'flavours' of FGRP3. You can read Bernd's response here. I have around 30 GPU endowed hosts running BRP5, with the CPUs in all of those doing FGRP2/FGRP3. There are a small number of those deliberately set up to help clean up the dregs of FGRP2 and so far, they are managing to keep up a cache of FGRP2 resends. Of course, pretty soon now, that will completely dry up.
When the GPU version of FGRP3 reappears, I anticipate a number of problems. I have a number of hosts with HD7850 GPUs running BRP5 tasks 4x. This automatically reserves 2 cores from a quad core host and if I reserve a third core, there is a small improvement in GPU performance. My understanding is that each FGRP3 GPU task will grab a full CPU core so if I try to run them 4x, there'll be no CPU processing at all. Also, I'm actually running a lot of GTX650 GPUs on dual core hosts with all cpus running FGRP3 and up to 3x on the GPU. That's certainly not going to fly when FGRP3 GPU tasks enter the mix.
I think the most likely thing I will have to do when FGRP3 GPU tasks reappear, is shift all my GPU hosts into a particular venue and deselect FGRP3 for that venue entirely. The GPUs will continue with BRP5 and the CPUs will be put to work on a totally different project that doesn't have a GPU app. Sure, I'll experiment with at least one host to try out FGRP3 GPU tasks but I'm guessing I wont be particularly happy with the expected loss of efficiency. Maybe I'm being too pessimistic :-).
Cheers,
Gary.
RE: Thanks, I'll try
)
I'd be very surprised (but extremely thankful) if that worked :-).
Theoretically, setting zero would attempt to run an infinite number of tasks simultaneously. I would guess that there would be some built in upper limit, so, for example, you might end up trying to run something like 10 concurrent tasks (ie factor of 0.1 or less). The other alternative would be to make 'impossible' factors just default to a value of 1 so you end up running just one at a time.
Also, I doubt that setting this factor to zero would stop the downloading of GPU tasks. It seems to me that the factor only controls how many crunch simultaneously so I can't see how it would stop task acquisition.
Of course, I could be totally wrong!
Cheers,
Gary.
RE: When the GPU version of
)
You could always override the project's default reservations with one of your own in a app_config.xml (and if you're running one of the latest Boinc dev versions, you can do it by app_versions too)
Claggy
Gary after several hours of
)
Gary after several hours of thought I reached the same conclusion as you presented, setting to zero should not work.