Intel® oneAPI Math Kernel Library
Ask questions and share information with other developers who use Intel® Math Kernel Library.
6956 Discussions

Linpack Benchmark on Windows7 and CentOS

Tomuro__Takahiko
Beginner
791 Views

I had run Intel Optimized LINPACK Benchmark on Windows 7 and CentOS 5.5.
Platform is DP server and CPU is X5680 (3.33GHz) and memory is 24GB.
(Benchmark is downloaded from here)
http://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download/

I had tested this benchmark without any VM and both is standalone OS environment. And BIOS setting is same.

Intel LINPACK data

Current date/time: Thu Aug 12 13:24:25 2010

CPU frequency: 3.333 GHz
Number of CPUs: 12
Number of threads: 12


Benchamrk number on Windows 7 is...

Size LDA Align. Average Maximal
10000 10000 4 103.7096 104.5793
15000 15000 4 108.3996 111.6880
20000 20000 4 132.7701 133.2039
25000 25000 4 106.5018 106.8729
30000 30000 4 136.6855 136.6855
35000 35000 4 109.1213 109.1213
40000 40000 4 139.4487 139.4487

Benchmark number on CentOS is...

Size LDA Align. Average Maximal
20000 20016 4 144.3745 144.5568
22000 22008 4 146.2399 146.3934
25000 25000 4 147.2634 147.2977
26000 26000 4 148.0729 148.1387
27000 27000 4 148.4799 148.4799
30000 30000 1 148.9899 148.9899
35000 35000 1 149.3800 149.3800
40000 40000 1 150.9887 150.9887
45000 45000 1 151.9700 151.9700


Performance number on Windows 7 is much lower than result on CentOS.
Do I need to check something to run on Winodws 7?

Thank you in advance..


0 Kudos
1 Reply
TimP
Honored Contributor III
791 Views
You didn't mention your KMP_AFFINITY setting. It's more critical on Windows, but may help linux also.
Running with HT enabled may be problematic, even when you set 1 thread per core.
0 Kudos
Reply