GPU utilization factor of BRP apps
Anonymous
6 Aug 2013 17:19:00 UTC
Topic 197120
(moderation:
if one plugs in a value of 0.5 into this field this translates into: gpu usage 0.5
cpu usage 0.2
how does one compute these values from the value entered on the E&H "HOME Preferences". The reason I ask is because I support S*H and if I change the value on the E&H website I should enter equivalent values in the Seti projects app_config.xml file on my PC.
Also how does one compute from the entered value how many GPU work units will be run?, i.e., .5 correlates to two jobs. How many is .4?
Is there a spreadsheet? :>)
GPU utilization factor of BRP apps
)
The utilization factor only controls how much of a GPU the task claims it will use, it does not change the CPU part.
So it goes like this:
[pre]Number of task Utilization factor Value shown in Boinc
1 1 0.2 CPU + 1 GPU
2 0.5 0.2 GPU + 0.5 GPU
3 0.33 0.2 CPU + 0.33 GPU
4 0.25 0.2 CPU + 0.25 GPU[/pre]
And so on...
Essentially you tell Boinc that one task won't fully utilize the GPU so it's ok to start more. Boinc will then start tasks so that the sum of GPU usage will be 1 or as many as will fit without going over 1. The utilization factor don't have to be the exact numbers as shown above, you could fill in 0.51 for Einstein and 0.49 for Seti to get either one Einstein task and one Seti task together or 2 Seti tasks together but not 2 Einstein.
To change the CPU part you'll have to go for a app_config.xml file but that's normally not needed here.
so would the following be
)
so would the following be true if I wanted to run 7 tasks for E&H only:
utilization factor = 1 / number of desired tasks
or 0.14
and would the 0.2 cpu value remain unchanged?
Yes, 0.14 x 7 = 0.98 so Boinc
)
Yes, 0.14 x 7 = 0.98 so Boinc would start 7 GPU tasks.
The 0.2 CPU will still be the same for every task and would add up to 0.2 x 7 = 1.4 so Boinc will reserve one CPU core for these tasks.
But before you fire up 7 parallel tasks start with just the one and monitor the GPU usage, then add one more task and monitor the usage again, then add another and so on.
What your looking for is when you reach a GPU usage of 90% or more and adding another task don't raise the usage. Then monitor the completion times and maybe try with one more and one less to see what gives the best performance on your particular setup.
I'm running x2 on my GTX660Ti and x1 on the Intel HD 4000 plus 5 task on the CPU. Any more CPU task and the GPU performance goes down. I reckon it's some sort of bottleneck in the memory subsystem.
RE: Yes, 0.14 x 7 = 0.98 so
)
Understood. I am currently running two GPU tasks for both E&H and S&H with a GPU temp of 53C. I believe that I can push the 650 ti a wee bit more and be safe. My remaining question is when trying different values for E&H should I also change the app_config.xml file under the seti project on my pc to reflect the same values?
I really do appreciate you taking the time to explain this. I feel that your responses will help clarify how this attribute functions within E&H and other projects and will be of great help to others new to distributed computing.
Thanks again.
The GPU utilization factor
)
The GPU utilization factor set in Einstein will have no effect on S@H, if you want to keep them in sync that is up to you. I doubt you will improve your scores going above 2 concurrent tasks as the processing bottleneck is memory bandwidth and not the amount of memory available.
--------------------------
- Crunch, Crunch, Crunch -
--------------------------
RE: My remaining question
)
Not necessarily, it might be that Seti runs best on an different number of simultaneous tasks and a mix of the projects needs yet another. Can't really say as I mostly just run Einstein.
Thank you for your kind words and I'm happy I could help! =)
RE: RE: Yes, 0.14 x 7 =
)
You won't have any trouble running your 650Ti at 53C in fact 63C will still run with no problems (I have a couple of them running OC'd)
Thanks to everyone who
)
Thanks to everyone who responded. More things to ponder.
I've had a GTX 650 hit over
)
I've had a GTX 650 hit over 72C without developing any errors.
No AC, poor case airflow, and so on.
It probably does not help the longevity of course, but the maximum temp listed is 98C. Geforce says the Ti has even more leeway (105C). So I tend not to worry about a few degrees here and there.
It is best/easier to use
)
It is best/easier to use app_config.xml method rather than going thru the Project-Preferences page (leave it at 1.0).
You can modify the XML file & issue "Advanced -> Read Config File" to immediately see results.
The Project-Preferences method on the other hand will only update BOINC-client when new tasks are issued (e.g. next batch).
terencewee*
Sicituradastra.