Firefly and PC GAMESS-related discussion club


 
Learn how to ask questions correctly  
 
 
We are NATO-free zone
 



Re^3: Firefly's parallel calculation: can't get fully CPU load

Alex Granovsky
gran@classic.chem.msu.su


Hi,

it seems you was using a very old version of the PC GAMESS.

Yo do not have four cores. You only have two physical cores
each with two logical processes.

You should read this document:

http://classic.chem.msu.su/gran/gamess/smp.html

By default, Firefly binds itself to the every second logical core
of a same physical core. This usually gives the best possible
performance. Some types of jobs benefit of the use of all logical
processors in parallel jobs. For these jobs, one needs to provide
additional input:

 $smp httfix=.f. $end



Kind regards,
Alex Granovsky


On Mon Mar 18 '13 2:40pm, Alexey wrote
--------------------------------------
>The result of FIREFLY-8 is the following:
>1 thread  - 25% CPU load (total 25)
>2 threads - 25+25% CPU load (total 50)
>3 threads - 25+12+13% CPU load (total 50)
>4 threads - 25+8+8+9% CPU load (total 50)

>Its output file contains the following text from the start of the file:
>
>============================================================
>                   Grand Master is running on host MAIN
>                   rank #     1 is running on host MAIN
>                   rank #     2 is running on host MAIN
>                   rank #     3 is running on host MAIN

>...
>[title block, framed with asteriks, is skipped]
>...

> Core i7    / Win32  Firefly version running under Windows NT
> Running on Intel CPU:  Brand ID  0, Family  6, Model  42, Stepping  7
> CPU Brand String    :  Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz        
> CPU Features        :  CMOV, MMX, SSE, SSE2, SSE3, SSSE3, SSE4.1, SSE4.2, AVX, HTT, MWAIT, EM64T
> Data cache size     :  L1 32 KB, L2  256 KB, L3  3072 KB
> max    # of   cores/package :   8
> max    # of threads/package :  16
> max     cache sharing level :  16
> actual # of   cores/package :   2
> actual # of threads/package :   4
> actual # of threads/core    :   2
> Operating System successfully passed SSE support test.
> Operating System supports AVX/FMA.
>
>
> PARALLEL VERSION (UNIFIED) RUNNING USING    4 PROCESSES (NODES)

> EXECUTION OF FIREFLY BEGUN 15:21:19 18-MAR-2013    

>            ECHO OF THE FIRST FEW INPUT CARDS -
>...
>[skipped]
>...
>   99000000 WORDS OF MEMORY AVAILABLE

> This job is executing on     1 unique host(s)
> Minimum number of processes per host is:    4
> Maximum number of processes per host is:    4

> Warning: HTT is enabled, bitmask of physically unique cores is 0x0000000A

> SMT aware parts of program will use              2 threads.

> Creating thread pool to serve up to            128 threads.

> Activating Call64 option.

> Using 64-bit DGEMM by default.
>
>
> Loading P2P interface library... loaded successfully (version 1.9).
> Initializing global P2P interface... topology done.
>============================================================
>
>
>
>Version of PC-GAMESS is 6, June 1999. Its output file starts with this:
>
>============================================================
> Pentium    / Win32  PC GAMESS version running under Windows NT
> Running on Intel CPU:  Brand ID  0, Family  6, Model 10, Stepping  7
> CPU Brand String    :  Intel(R) Core(TM) i3-2100 CPU @ 3.10GHz        
> CPU Features        :  CMOV, MMX, SSE, SSE2, SSE3, SSE4, EM64T                
> Data cache size     :  L1 32 KB, L2  256 KB, L3 3072 KB
> # of cores/package  :  8
> Operating System successfully passed SSE support test.

> Warning: this PC GAMESS version was optimized for a different CPU type!
> The overall performance degradation is expected.
> For better performance, use Pentium III optimized PC GAMESS version.
>
>
> PARALLEL VERSION (UNIFIED) RUNNING WITH   4 NODES    
>============================================================
>

[ This message was edited on Fri Mar 22 '13 at 6:30pm by the author ]


[ Previous ] [ Next ] [ Index ]           Fri Mar 22 '13 6:30pm
[ Reply ] [ Edit ] [ Delete ]           This message read 969 times