question about the different "work" on einstein@home
7 Aug 2012 10:38:27 UTC
Topic 196473
(moderation:
hi,
if this post is in the wrong subforum fell free to move it :).
i just started einstein@home again yesterday. i got my first batch of work and crunch it down. while seeing the diffrent work inside boinc manager i came up with some questions, which hopefully someone can answer me.
im getting BRPS data for cuda and for cpu cores.
and also i get gravitational wave workunits which are done by cpucores.
my pc setup does the BPRS data for cuda in the range of 40-45 mins per workunit. while one cpu core do such a workunit in more then 17 hours.
isn't this quite a bit borderline to give me cpu work which take 17 hours, when my gfx card does the same win just 45 minutes ?
wouldnt it way more efficient to give me BPRS work for cuda only and for the cpu cores just work for the gravitational wave ?
this post isn't meant as a rant about speed or stuff as i simply don'T care yout credits and stuff.
i just have hard time to understand why 4 cpu cores are locked with data for 17 hours when there is other work like s6lv for example which is cpu only for technical reasons.
if you calculate this behaviour up for like 1000 pc's. it looks for me like i giant waste of electrical energy to be honest.
question about the different "work" on einstein@home
)
Hi,
welcome (back) to E@H.
Goto http://einstein.phys.uwm.edu/prefs.php?subset=project
and set "Run CPU versions of applications for which GPU versions are available" to "No".
Michael
edit: You might want to set "GPU utilization factor of BRP apps" to "0.5", so two BRP tasks run in parallel.
Team Linux Users Everywhere
Hi agony, I'm sure the
)
Hi agony,
I'm sure the project staff are pleased to welcome you back.
As Michael points out, the project preferences are designed to give you the sort of control that you seek and the ability to contribute as efficiently as possible. You should have a good read through all of the preferences (both general computing and project specific) so that you get a good idea of how you may configure things to your liking. If you are not sure of what exactly a specific preference does, please ask and someone will fill you in.
If you take Michael's advice and set the preference to run multiple GPU tasks simultaneously, please realise that you will further improve efficiency by reserving a CPU core to support the GPU crunching. GPU tasks do need significant CPU support and the easiest way to provide that support would be to set the computing preference for "On multiprocessors, use at most ..." to 75% so that one core is prevented from running a 'CPU only' job. That will further speed up the tasks running simultaneously on the GPU. With that setup you should see 3 CPU tasks and 2 GPU tasks running simultaneously.
If you have more than 1GB GPU RAM, you could even consider more than 2 simultaneous GPU tasks and in that case you may get better results again, particularly if you allow two free CPU cores in support. In recent months, a number of people have posted their experiences with multiple tasks on both AMD/ATI and NVIDIA GPUs. Posts are in a number of places but take a look at this particular thread to see where performance information for a large number of different GPUs has been gathered in one place.
When you change your website preference settings to prevent BRP tasks from being sent for crunching by a CPU core, you need to 'update' your BOINC client to force it to contact the project and be told of the change immediately. For any unstarted BRP CPU tasks you already have in your cache, you should consider aborting them immediately to save wasting any CPU time on them. I'd probably also consider aborting ones that have started if they haven't made much progress. That's entirely up to you.
Cheers,
Gary.
hi again, the change in
)
hi again,
the change in the preferences did the trick. and einstein now does what i want :)
thanks again for the help and sorry for not checking the preferences to find this out by myself