[CPMD-list] Re: TH

[Axel Kohlmeyer]

On Tue, 1 Jul 2003 16:58:08 -0500  "Weiguo Yin" wrote:

> 
> Dear all,
> 
> >as far as i remember the only advice for this problem on an SGI so
> >far was to compile for 32 bit, i.e. get rid of -DPOINTER8,
> >use -n32 instead of -64, and recompile everything.
> >
> 
> It works for cpmd-test/std/Si64. Thanks to Axel kohlmeyer. My SGI (CPU:
> MIPS R12000 + FPU: MIPS R12010) did not crash when SPLINE POINTS were set

hi,

my interest was in varying 'SPLINE POINTS' _without_ recompiling.

> 500,1000,3000,5000, and 10000. Besides, neither using v3.5.4 nor
> compilation without -r8 has significant impact on the speed of this simple
> test.

that was on a linux pc, which most likely allocates memory differently,
so you won't necessarily see an effect. i suspect a silent memory corruption
by crossing boundaries of dynamically allocted memory blocks.

> 
>                     ****************************
>                     **    FURTHER QUESTION    **
>                     ****************************
> Why are "THE NEW SIZE OF THE PROGRAM" and "MEMORY" obtained in the test on
> SGI much larger than those given in cpmd-test/std/Si64/out-1 and out-2?
> Please refer to the following data:

because the example job was run on a parallel machine with 32 cpus.
cpmd uses a distributed memory parallelization. have a closer
at the PARAPARAPARAPARAPARA section. it tells you how the job
is distributed over the nodes.

> 
> -------------------------- cpmd-test/std/Si64/inp-1
> ------------------------
> [My result, 2 processors, -n32 -r8]
>   ***    RWFOPT| THE NEW SIZE OF THE PROGRAM IS  76611 kBYTES ***
>   [PEAK NUMBER   83]      PEAK MEMORY      4806098 =   38.4 MBytes
> 
> [cpmd-test/std/Si64/out-1, 32 processors]
>   ***    RWFOPT| THE NEW SIZE OF THE PROGRAM IS   8080 KBytes ***
>   [PEAK NUMBER   82]      PEAK MEMORY       683044 =    5.5 MBytes
> 
> -------------------------cpmd-test/std/Si64/inp-2---------------------------
> [My result, 2 processors -n32 -r8]
>   ***    GMOPTS| THE NEW SIZE OF THE PROGRAM IS 67846 kBYTES ***
>   ***     PHFAC| THE NEW SIZE OF THE PROGRAM IS 70139 kBYTES ***
>   [PEAK NUMBER   95]      PEAK MEMORY      4185128 =   33.5 MBytes
> 
> [cpmd-test/std/Si64/out-2, 32 processors]
>   ***    GMOPTS| THE NEW SIZE OF THE PROGRAM IS  5324 KBytes ***
>   ***     PHFAC| THE NEW SIZE OF THE PROGRAM IS  5524 KBytes ***
>   [PEAK NUMBER   95]      PEAK MEMORY       629261 =    5.0 MBytes
> 
> It appears that CPMD runs on the SGI ORIGIN platform are inefficient. Could
> somebody tell me how to tune CPMD? Thanks in advance.

you cannot talk about efficieny with such small runs (look at the cutoff!).
for bigger jobs memory usage seems pretty invariant between platforms.
this is to be expected, since most of the memory is needed for real*8 data.

regards,
	axel kohlmeyer.

> 
> Weiguo Yin
> 
> 

--

=======================================================================
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
=======================================================================
If you make something idiot-proof, the universe creates a better idiot.