Why Are Benchmark Results Lower Under Linux ??

Message boards : Number crunching : Why Are Benchmark Results Lower Under Linux ??
Message board moderation

To post messages, you must log in.

AuthorMessage
Terror Australis
Volunteer tester

Send message
Joined: 14 Feb 04
Posts: 1806
Credit: 256,446,391
RAC: 11,722
Australia
Message 722288 - Posted: 5 Mar 2008, 12:51:31 UTC

Hi All
I have noticed that for exactly the same machine (Proc. Memory, MoBo etc.) that when running Linux, the benchmark scores are only about half the benchmark scores under Windows. I noticed my Linux boxes have a Result Duration Correction Factor of 0.2 to 0.45 and the Windows boxes go from 0.7 to 0.9

This makes no difference at all to the processing times and usually Linux will run a higher RAC than Windows on the same machine.

I'm just wondering why the BOINC Benchmark test is so inconsistant between the two OS's ?

Regards
Brodo
(Getting right into the Tech of things :-)
ID: 722288 · Report as offensive
Nick: ID 666
Volunteer tester

Send message
Joined: 18 May 99
Posts: 12804
Credit: 33,274,372
RAC: 19,094
United Kingdom
Message 722332 - Posted: 5 Mar 2008, 15:30:14 UTC

Don't think anybody has found a reason for it yet. You have discovered why Seti moved to counting Flops to calculate credits.
If you had crunched units with Linux as the OS when it was Benchmark * time, as benchmark was low and time quicker on Linux the claimed was very low. Two Linux machines, in the quorum, on one unit and your RAC went done the pan.
ID: 722332 · Report as offensive
1mp0£173
Volunteer tester

Send message
Joined: 3 Apr 99
Posts: 8423
Credit: 356,897
RAC: 0
United States
Message 722373 - Posted: 5 Mar 2008, 17:11:55 UTC - in response to Message 722288.  

Hi All
I have noticed that for exactly the same machine (Proc. Memory, MoBo etc.) that when running Linux, the benchmark scores are only about half the benchmark scores under Windows. I noticed my Linux boxes have a Result Duration Correction Factor of 0.2 to 0.45 and the Windows boxes go from 0.7 to 0.9

This makes no difference at all to the processing times and usually Linux will run a higher RAC than Windows on the same machine.

I'm just wondering why the BOINC Benchmark test is so inconsistant between the two OS's ?

Regards
Brodo
(Getting right into the Tech of things :-)

Most likely different compilers.

The duration correction factor takes care of that (among other things).
ID: 722373 · Report as offensive
Profile The Gas Giant
Volunteer tester
Avatar

Send message
Joined: 22 Nov 01
Posts: 1904
Credit: 2,646,654
RAC: 0
Australia
Message 722394 - Posted: 5 Mar 2008, 19:24:49 UTC

It's been known for many years that Linux benchmarks lower than 'doze. On my boxes Linux bm's at about 85% of 'doze. At one point in time Rom released a BOINC client that bm'd at about 105% of 'doze after weeks of discussions in the alpha mailing list. He soon reverted back however, after the version of a library was found to be not as backwards compitable as originally indicated and older linux versions would not work with this new build. Windoze appears to do backwards compitability better......

Live long and BOINC!

Paul
(S@H1 8888)
And proud of it!
ID: 722394 · Report as offensive
Profile Toby
Volunteer tester
Avatar

Send message
Joined: 26 Oct 00
Posts: 1005
Credit: 6,366,949
RAC: 0
United States
Message 722397 - Posted: 5 Mar 2008, 19:57:57 UTC

Its all about compiler optimizations. By dfault the linux client is less optimized than the windows one. But if you are using Gentoo like I am on some of my systems then the boinc client is compiled from source with whatever optimizations you have set for the whole system. I recently changed one of my 4400+s from windows to linux. The old windows benchmark was 2246.68 floating point and 4151.96 integer. The same hardware, now running linux scored 2173.97 floating point and 5954.82 integer. As you can see the floating point is slightly lower (under 10%) and the integer benchmark is actually higher in linux. But this is just because it was compiled with different optimizations turned on. If I used the stock BOINC client it would be lower.

I may delete the windows host soon but until then you can compare them side by side
A member of The Knights Who Say NI!
For rankings, history graphs and more, check out:
My BOINC stats site
ID: 722397 · Report as offensive

Message boards : Number crunching : Why Are Benchmark Results Lower Under Linux ??


 
©2019 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.