Advanced search

Message boards : Number crunching : CPU time vs. Run time

Author Message
jjwhalen
Send message
Joined: 23 Nov 09
Posts: 29
Credit: 17,591,899
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 14815 - Posted: 30 Jan 2010 | 0:44:52 UTC

I'm curious. What is it about the acemd application (or the WUs) that causes such a wide variation in the amount of CPU time consumed by processing a result using what is primarily a GPU application?

For example, my last 2 completed results had ratios of CPU to Run time of 6,268.40 to 45,519.33 and 3,107.15 to 47,449.10 respectively, which makes one's ratio twice as much as the other. If both results ran uninterrupted (yes), on the same hardware (also yes), and the host's quad CPU cores are used almost exclusively for BOINC (which they are), this big a difference seems "interesting." And this is just one example.

Once again I'm not complaining, just curious.
____________

Profile robertmiles
Send message
Joined: 16 Apr 09
Posts: 502
Credit: 580,600,933
RAC: 119,102
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 14830 - Posted: 30 Jan 2010 | 15:17:12 UTC
Last modified: 30 Jan 2010 | 15:17:54 UTC

At least for one of the recent beta acemd versions, enough more CPU time is used than expected that I'd expect the ratio of GPU capability to CPU capability to have more of an effect than usual - in other words, that version is no longer primarily GPU. You might want to watch for signs that this applies to some of the other acemd versions as well.

Snow Crash
Send message
Joined: 4 Apr 09
Posts: 450
Credit: 539,316,349
RAC: 0
Level
Lys
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 14836 - Posted: 30 Jan 2010 | 16:28:01 UTC - in response to Message 14830.

The people here at GPUGrid are very smart. They didn't just use someone else's scientific software (like an autodoc package) but they invented their own! ACEMD allows them to run many different types of experiments which have a wide variety of parameters. If you take a look at the science page you will get an idea of the diverse possibilities they have created a framework for. So yes, there are very different computing times and CPU utilizations but the bigger picture is they have opened up a whole new avenue of molecular research.
____________
Thanks - Steve

Post to thread

Message boards : Number crunching : CPU time vs. Run time

//