BRP CUDA set to .33 but not working!
21 Sep 2012 20:16:17 UTC
Topic 196532
(moderation:
I have set the BRP to 0.33 but still only get:
(0.2 CPU"s + 1 NIVIDIA GPU (device 0))
(0.2 CPU"s + 1 NIVIDIA GPU (device 1))
I should be getting:
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 0))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 0))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 0))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 1))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 1))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 1))
Does anyone have an idea how to get this to work?
I am running an AMD 8 core with 2-GTX 460 video cards.
Seti@Home works fine but Einstien@home does not.
Thanks,
John
BRP CUDA set to .33 but not working!
)
John,
I think the change only affects work units sent after your change. Any work units already on your system will not be affected. Be patient and wait for new work units.
Joe
Hmmmm! This should not be the
)
Hmmmm! This should not be the case but thank you for the info. I will wait a few days to see if there are any changes.
Running Seti@Home changes can be made anytime. Once you click on update you will see your changes.
@ Any Rate Thanks,
john
If it is a setting that
)
If it is a setting that changes the tasks themselves and not the setting on your computer, the change wouldn't be immediate on SETI either. ;)
This parameter is not a
)
This parameter is not a standard BOINC parameter an then its not updated in the standard way, so doing a manual update may not work.
It will be updated in your hosts the next time that they get new BRP4 work.
Once the parameter is read it will be applied to all the running tasks no matter when they were downloaded.
If you had several day of cache and you change the factor (to run less concurrent WUs) then the cache may become suddenly overfilled and reverting to the original factor might take some time because it wont be asking for more BRP WUs for a while...
RE: This parameter is not a
)
Right, at SETI you have to change parameters in app_info.xml file, which
will be created if you use the LUNATICs V40 Installer.
Or have the Optimized Applications installed manually, but then you'll
have write an app_info.xml file yourself.
In Einstein, I think it's a 'parameter' in the science app., which is changeble.
In fact you 'instruct' the science app. to use 1 a half or a third of the
Compute Units/CUDA cores of the GPU(s).
I have set the BRP to 0.33
)
I have set the BRP to 0.33 and I am now getting the following results with new work units. My next question is how do I change the '0.2 CPU's' to something else like '0.1'?
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 0))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 0))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 0))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 1))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 1))
(0.2 CPU"s + 0.33 NIVIDIA GPU (device 1))
Thanks,
John
RE: ... My next question is
)
You could only do that through a customised app_info.xml.
But why would you want to do that? If you look at your current results, you can see that the CPU time being used is actually around 20% of the run time anyway. If you were to force it to something less, you would only bottleneck the GPUs even further.
Some questions about your setup:
1. How many of your 8 cores are actually crunching?
2. Do you use preferences to confine BOINC to less than 100%? If so, what have you set?
3. Are both your PCIe slots x16 slots?
4. Do you realise that running x3 tasks on 2 GPUs each of which has only 1GB RAM is risking running out of memory type failures and is also likely to give lower performance than running x2 tasks on those GPUs? Have you thoroughly tested out a X2 setup where 4 X 0.2 CPUs would actually be less than 1 CPU and you wouldn't need to think about trying to lower the 0.2 CPUs parameter?
I would be surprised (particularly with an AMD processor) if you could make a x3 tasks on 2 GPUs setup work better than x2 tasks. I imagine you would need at least 2 free cores and probably more.
Cheers,
Gary.
RE: RE: ... My next
)
Setting it to .1 or .5 of a
)
Setting it to .1 or .5 of a CPU means nothing. Its only an estimate so BOINC can schedule tasks. It will actually use whatever % of CPU time it wants regardless of the setting.
If you had 5 tasks running and the estimate was .2 then BOINC needs to commit a whole CPU core, in which case it would reduce the number of CPU tasks. Depending on your BOINC version it will allow a certain amount of over-commiting of the CPU.
In the case of Seti their app has been more optimised which is why it has a lower CPU estimate. If you look at your results and compare the Seti CPU time with Einstein CPU time you will see Seti uses much less, especially if you are using the x41g or x41z optimised apps.
BOINC blog
RE: Setting it to .1 or .5
)
ok so the application is auto set, 8 cores using 0.2 of each cpu and 6 cuda cores or gpu's running at 0.33 for a total of 14 wu's at a time. I can live with that. If I change Einstein@home to 0.5 gpu's than I believe that will effect Seti@home as well making it also run 0.5 gpu's. I'll have to watch that to see what happens latter when I change things up. It would be nice to run Seti@Home on one computer and Einstein@Home on an other computer with out having to set up 2 accounts to run then independently. Just thinking out loud!
Thanks,
John