Robert Carter's question about CPU Efficiency
9 Nov 2005 7:48:24 UTC
Topic 190139
(moderation:
I recently joined Einstein at Home with two computers. There is about a 30% difference in the floating point and integer speeds between the two computers, and the 'seconds of CPU time' of the fast machine is about half of the other one. The faster machine is completing about seven times as many work pieces per day however. The explanation seems to be in the "Average CPU efficiency" number which is .37 on the slower machine, and.99 on the other. How is the "Average CPU efficiency" number in the "Computer Summary" page generated? Thanks, Bob
Cheers,
Gary.
RE: I recently joined
)
Hi Bob,
Thank you for joining the EAH project!! It's good to have you on board.
I shifted your post here as this is a much more suitable place to get questions like this answered.
You are correct in assuming that Average CPU efficiency is the key parameter to look at for an answer. What it basically means is that your slower machine is only giving 37% of its total cpu cycles to EAH and 63% are going to other processes. However in this case there might be another answer. It's listed as a mobile celeron so I would imagine it's a laptop. Does it have any sort of thermal throttling? Does it reduce its clock speed when there is no keyboard activity? If you have it connected to the mains but running on a "conserve power" type setting, maybe it's doing just that and maybe this is causing BOINC to think that it's a lot slower than the benchmarks suggest?
Have a think about these sort of things and see if you can work out what's happening. I imagine if there really were another process eating up 63% of your CPU then you would probably know about it :).
Cheers,
Gary.
RE: I'm creating this
)
Gary,
I don't have an Intel, so I thought I'd ask you this: Do you have a P4 machine handy (running 5.x), to check if they're inherently running lower CPU efficiency? Because it they ordinarily go at 90+%, I'm guessing that Robert's Intel machine is either running something else that's very CPU-intensive, or that Boinc is often suspended but not closed (or Manager closed but not Daemon closed from System Tray). I believe that either of these scenarios would account for the low efficiency #s.
Regards,
Michael
microcraft
"The arc of history is long, but it bends toward justice" - MLK
Michael, I have a PIII
)
Michael,
I have a PIII which started crunching a week ago on a 5.2.X BOINC. Its current value is 0.99518.
I have a P4 which started on 4.19 back in February but has been upgraded to 5.2.X about a week ago. Its current value is 0.99952.
As long as the OS is 2K or XP, (as the above are) I think the values are accurate.
I have a PIII running Me with the same history as the P4 above. Its current value is 1.087969.
I have a Celeron running 98SE+unofficial patchkit with the same history as the PIII/Me and its value is 1.573784. WOW!! 157% of the available cycles!! I always knew that patchkit was hot stuff :).
Seriously, Windows 9x systems just don't handle apps being preempted and left in memory all that well I think. Things got a lot better when Microsoft "borrowed" some ideas from unix :).
Getting back to Bob's problem - I just don't believe that his laptop could be running something else that cpu intensive without him knowing about it. I also don't think it's anything to do with the daemon running but not the Manager. I'm actually experimenting with having the daemon running and then firing up the Manager just to see what's going on. When the Manager is completely closed, the daemon continues on at full speed with nothing to steal cpu cycles. Seems to work very well.
Cheers,
Gary.
Gary, Yes, I'm still
)
Gary,
Yes, I'm still awake. My thinking is that maybe with the daemon running but no manager to order crunching, that time would rack up without fulltime crunching, lessening efficiency. Other thing is maybe preference to work fulltime, but some OpenGL-based screensaver. Just tossing around ideas here, speculation.
Regards,
Michael
(edit) When I was testing one of the 5.x clients (can't remember which, 5.2.1, I think), I accidentally left the project suspended for a few hours, and the next time I looked at my summary page, the efficiency had dropped from the familiar .99xx to a value closer to .90. I didn't keep an eagle eye on it, but I don't think that drop was coincidence.
microcraft
"The arc of history is long, but it bends toward justice" - MLK