Advanced search

Message boards : Number crunching : How CPU perf is computed ?

Author Message
mickael_83
Send message
Joined: 25 Oct 09
Posts: 10
Credit: 12,911,036
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwat
Message 22881 - Posted: 3 Jan 2012 | 15:53:55 UTC

Hello Everybody !
Happy new year !

There is something that annoys me a little: if you take a look @ my computers, id 54975 & ID 114761, you can see they use the same CPU, Q6600 @ 2.4G.
Matador is overclocked @ 3.2.
If you look at the details, there is a huge difference in MIPS & Gflops.
Matador is @ 4Gflops and 16000 Mips, while the other is only @ 2.3Gflops and only at 4000 Mips.
even if matador is 25% faster on clock, I don't think it can make a so huge difference.

Do you think the OS can explain this ?

Profile skgiven
Volunteer moderator
Volunteer tester
Avatar
Send message
Joined: 23 Apr 09
Posts: 3968
Credit: 1,995,359,260
RAC: 0
Level
His
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 22918 - Posted: 8 Jan 2012 | 15:19:29 UTC - in response to Message 22881.
Last modified: 8 Jan 2012 | 15:26:50 UTC

The benchmark system is a Boinc feature, rather than a project issue. Linux and Windows tend to be somewhat unequal. IIRC Linux under-reports but generally does more work (on the CPU). GPUGrid is a GPU project so reported performance of the CPU is largely irrelevant to actual GPU task performance and credit. As some CPU is needed, the faster the CPU the better, but a generic benchmark does not accurately measure real performance on a real project.
____________
FAQ's

HOW TO:
- Opt out of Beta Tests
- Ask for Help

Profile Retvari Zoltan
Avatar
Send message
Joined: 20 Jan 09
Posts: 2343
Credit: 16,239,065,968
RAC: 3,161,193
Level
Trp
Scientific publications
watwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwatwat
Message 22974 - Posted: 14 Jan 2012 | 10:40:10 UTC - in response to Message 22881.

The benchmark of the 32 bit and the 64 bit BOINC manager are different. I have changed my WinXP from 32 bit to 64 bit recently, and the benchmark results of the 32 bit and the 64 bit BOINC manager are different even on the same hardware.
I don't exactly remember the numbers, but on the 64 bit client the floating point ops are lower by ~10%, and the integer MIPS are higher by ~40% on the same hardware.
I still have a 32 bit client, almost the same hardware as one of my 64 bit clients: (Core i7-870)
on the 64 bit client:
3623.97 MFLOPS
12621.2 MIPS
On the 32 bit client:
4200.95 MFLOPS
9044.98 MIPS

mickael_83
Send message
Joined: 25 Oct 09
Posts: 10
Credit: 12,911,036
RAC: 0
Level
Pro
Scientific publications
watwatwatwatwatwatwatwat
Message 22976 - Posted: 14 Jan 2012 | 13:05:58 UTC

Hello.
Yes i know it is a boinc feature, and GPUgrid does not really mind about cpu power, but i found this surprising.
The windows machine is 32 bits, while the gentoo machine is 64bits.
Having so much difference between two same processors caught my attention....
Thank you for your answers...

Post to thread

Message boards : Number crunching : How CPU perf is computed ?

//