This thread is locked.Only browsing is available.
Top Page > Browsing
Running openMX on Cray XC30
Date: 2014/01/28 23:58
Name: Yu Xie   <yxe@ornl.gov>

Dear Dr. T. Ozaki,

I've compiled the openMX 3.7 on Cray XC30 using the same optimization flag showed in the "runtestL.result_XC30" in the large_example directory. However, when I run the example. The speed of my compiled openMX is about half the speed of the reference one. For example the DIA512 case, the reference one used 37.93 sec, mine used 67.xx sec. Could you make some comments on how to make my own openMX to run as fast as the reference one since the platforms are the same.

Best,

Yu
メンテ
Page: [1]

Re: Running openMX on Cray XC30 ( No.1 )
Date: 2014/01/29 08:00
Name: T. Ozaki

Hi,

Cray XC30 I tested consists of Intel Xeon E5-2670 2.6GHz (Sandy-Bridge).
Does your XC30 have the same CPUs?
Also, did you use the same configuration (processes and threads) for the parallel
caculations?
Which lapack library did you use? (We used ACML.)

In fact, we struggled for much time to find out the best combination (compiler and
compiler options) so that the execution of OpenMX can be best efficient on our XC30.

Could you try a wide variety of combinations (compiler and compiler options)?
How about the cray compiler?

Good luck.

Regards,

TO
メンテ
Re: Running openMX on Cray XC30 ( No.2 )
Date: 2014/02/07 04:43
Name: Yu Xie  <yxe@ornl.gov>

Sorry for the late response, we have got the same speed as the reference XC30 with multi-threads. Thanks.
メンテ

Page: [1]