Boinc Benchmark values differ Linux vs Windows on same Hardware
29 Mar 2007 21:39:26 UTC
Topic 192556
(moderation:
Hello,
mostly I am crunching for two projects: E@H and QMC@H. E@H is done using the Windows client, QMC with Linux. Both on exactly the same ´hardware which is Intel Core2Duo E6600, 2GB RAM. As I understand it the core calculation is independent from the OS and the speed depends only on the Processor and RAM speed.
Even so the Benchmark result in the Boinc App gives me about 3200 Whetstone/6800 Dhrystone using Linux and 2200/4600 using Windows. (using just one of the two cores) both measured while boinc was the only program running (except for system processes). I did the benchmark several times for about a month and the difference is always in favor of Linux and always quite significant.
Could anyone help interpreting these results?
Thank you
Boinc Benchmark values differ Linux vs Windows on same Hardware
)
It's dependant on the compiler used, and level of optimization of the resultant executable.
I've read that the latest builds of BOINC have taken care of this problem for the most part.
HTH,
Alinator
Thanks. Do I have to update
)
Thanks.
Do I have to update something then or is it done automatically?
RE: Thanks. Do I have to
)
E@H and QMC@H both have fixed credits - only dependant of the WU and not of the benchmark...
Udo
Benchmarks are part of BOINC
)
Benchmarks are part of BOINC itself, so that you have to DL the latest version and install it.
Also, I'm not 100% sure if the latest recommended version (5.8.15) has those fixes in it, or if the conversation was talking about the current dev build.
Alinator
RE: RE: Thanks. Do I have
)
That's true but the benchmarks are used for more than just calculating credit.
Alinator
Currently 5.4.11 is installed
)
Currently 5.4.11 is installed in Windows and in Linux. So I will upgrade anyway.
I am not so familiar with the concept of Whetstone and Dhrystone so my guess was that there are multiply and add instructions and some such in the processor E6600 and using them gives you integer and floating point values which are independent from the OS. Like the measurement Instructions per second which is totally independent from the OS.
But this seems to be wrong?
I was looking around for the
)
I was looking around for the post I read where the BM performance on Linux was finally equivalent to Win boxes, but you think I can find it when I need it?
NOT! ;-)
Anyway, there are some other fixes in 5.8.15 so it's probably worthwhile doing the upgrade.
Alinator
And I heard it said that a
)
And I heard it said that a computer does not forget anything, :-)
Thanks anyway.
The improved benchmark is in
)
The improved benchmark is in the latest alpha build, not the release version.
I don't have a linux box, but I can say the alpha list has been quite quiet for the last few days. Nothing major has turned up.
Older distros might have a problem with it because of the version of glibc the compiler used. But give it a shot.
:)
Kathryn :o)
Einstein@Home Moderator
Thanks Kathryn, I knew I
)
Thanks Kathryn, I knew I hadn't imagined it! :-)
IIRC, I even saw reports where the BM's were actually a little higher than when running Windows.
Alinator