[CPMD-list] Help, compiling problem!

[Axel Kohlmeyer]

On Tue, 2 Aug 2005, Wone Keun Han wrote:

WKH> Dear  Shyma and Members,
WKH> 
WKH>  
WKH> 

WKH> The Linux seems to recognize the single cpu PC as two processor PC if
WKH> it has dual channel memory.
WKH> 
WKH> I mean LINUX-SMP can be installed.

you _still_ have only one processor. what is the point?
you would have two threads competing for the same cpu,
the same memory and the same floating-point unit(!).
on top of that, you have the overhead of compiling in
the thread management for openmp.

WKH> 
WKH> I am wondering if we can speed up CPMD by compiling it as SMP compared to
WKH> serial calculation.

based on the facts listed above, CPMD will become (much) slower.

in fact, i made quite a few tests on real multiprocessor PC hardware
and even there i found little evidence that suggested the use of
openmp. due to the external memory controller and the extremely
high clock rate, intel xeon cpu are worst, dual athlon (32-bit) are
a bit better, but in both cases the overhead for OpenMP is so high,
that you are either better off using two MPI threads per SMP node
or not using the second cpu AT ALL. thanks to the internal memory
controller opteron cpus are a different story, but again two MPI
threads is much more efficient than using OpenMP (unless you run across
are really, really largen number of nodes and even then i found a
higher speed gain using other tricks, e.g. TASKGROUPS than with 
OpenMP).

the best openmp performance on SMP machines, so far, i found on
with the power4/5 architecture (either on ibm hardware or on
mac osx machines using the ibm compiler). also using OpenMP is 
essential for many vector machines, but that is a different story.

WKH> 
WKH> If so, I would like to have more information.

hope this will satisfy you. you can find some (pretty old by now)
benchmark numbers and discussions on 
http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/cpmd-bench.html

regards,
	axel.

WKH> 
WKH>  
WKH> 
WKH> Thanks,
WKH> 
WKH>  
WKH> 
WKH> Wone Keun Han
WKH> 
WKH>  
WKH> 
WKH> 

-- 

=======================================================================
Dr. Axel Kohlmeyer   e-mail: axel.kohlmeyer at theochem.ruhr-uni-bochum.de
Lehrstuhl fuer Theoretische Chemie          Phone: ++49 (0)234/32-26673
Ruhr-Universitaet Bochum - NC 03/53         Fax:   ++49 (0)234/32-14045
D-44780 Bochum  http://www.theochem.ruhr-uni-bochum.de/~axel.kohlmeyer/
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.