Page 1 of 1
Project 3064
Posted: Tue Apr 22, 2008 12:53 am
by toaster8
Protein: p3064_lambda5_2003Extra SSE boost OK.
This project comes up as worth 1753 points and yet takes twice as long to run as a project 2605 worth 1760 points. Increase is warning me that it may not complete in time as it takes 44 hours at 38pph. It will definetly not make it for the preferred time. I am running an 8-core macpro and most other wu's complete no problem. I am guessing that there is either a problem with this project, it was given too low of a point score or my machine just does not like this one.
Re: Project 3064
Posted: Tue Apr 22, 2008 8:27 am
by zorzyk
toaster8 wrote:Protein: p3064_lambda5_2003Extra SSE boost OK.
This project comes up as worth 1753 points and yet takes twice as long to run as a project 2605 worth 1760 points. Increase is warning me that it may not complete in time as it takes 44 hours at 38pph. It will definetly not make it for the preferred time. I am running an 8-core macpro and most other wu's complete no problem. I am guessing that there is either a problem with this project, it was given too low of a point score or my machine just does not like this one.
I don't know MacPro with 8 cores, but on Q6600/XP I get the following results (depending on CPU speed):
Code: Select all
Project 3064 - 1 x SMP:
Q6600 @ 2.93 GHz 16h 00min 2629 PPD 164 PPH 9:36 TPF
Q6600 @ 3.18 GHz 14h 47min 2846 PPD 193 PPH 8:52 TPF
I can't compare to 2605 because I didn't fold it.
How are your results comparing to other people who fold on MacPro/8 cores?
Re: Project 3064
Posted: Tue Apr 22, 2008 9:07 am
by toTOW
zorzyk wrote:I can't compare to 2605 because I didn't fold it.
You can compare with the 2653 ... they fold at the same speed but 2605 is linux only
Re: Project 3064
Posted: Tue Apr 22, 2008 9:28 am
by zorzyk
toTOW wrote:You can compare with the 2653 ... they fold at the same speed but 2605 is linux only
OK
Project 3064 gets almost exactly the same results as 2653, but 2653 folds slightly quicker (1-2 seconds quicker per frame).
Re: Project 3064
Posted: Tue Apr 22, 2008 9:36 am
by toTOW
Yes 30xx are generally a little bit slower than 26xx used to
Re: Project 3064
Posted: Tue Apr 22, 2008 2:57 pm
by toaster8
Slightly slower I could handle but this is less than half the speed. Is this normal?
Re: Project 3064
Posted: Tue Apr 22, 2008 3:04 pm
by zorzyk
In my opinion two times slower is not normal. MacPro 8 cores is built on double Xeon quad core 2.8MHz basis.
Two times slower means that your 3064 is utilizing 2 cores instead of four. How do you run your SMP client?
Re: Project 3064
Posted: Tue Apr 22, 2008 3:41 pm
by bruce
zorzyk wrote:In my opinion two times slower is not normal. MacPro 8 cores is built on double Xeon quad core 2.8MHz basis.
Two times slower means that your 3064 is utilizing 2 cores instead of four. How do you run your SMP client?
Nobody is talking about the cache size and how the caches are shared between the various CPU-cores, compared to the same information on the benchmark machine. These can be much more important than the simple number of CPU-cores on your hardware. Tightly coupled CPUs are vastly superior to ones with a much slower data rate between CPUs . . .
on specific WUs. On other WUs, it's not so important.
FAH will use whatever resources you provide, and one of the drawbacks of any benchmarking system is that when your hardware is different than the benchmark machine, any of the differences might be important, not just one single number.
I have a double-Xeon machine, and the data rate between the cores that share cache is vastly faster than the data rate between cores that are on different chips. In some cases it acts like a quad, and in other cases it acts much like a pair of independent dual processors.
Re: Project 3064
Posted: Sun May 24, 2009 4:26 pm
by Outback_Jon
Guess I'm about a month late to this thread, but I just recently got a p3064 on my dual-core Opteron @ 2.5GHz. My PPD is good, but I'm not going to make the preferred deadline. This one is going to take a bit over 2 days, even if I don't use my computer for anything else. And the deadline is 1.8 days.
EDIT: p3065 gives similar results. I'll make the final deadline, but not the preferred.