[gmx-users] Re: IBM p-series performance

Pim Schravendijk schraven at mpip-mainz.mpg.de
Wed Oct 20 14:34:07 CEST 2004


I had similar results. 

Searching on power4 in the user-list will show you some recent benchmarks: 

http://www.gromacs.org/pipermail/gmx-users/2004-September/012083.html

About the cause of this and any hope in getting a better performance (not 
much chance though :( ), see:

http://www.gromacs.org/pipermail/gmx-users/2004-August/011655.html
(a reaction on:
http://www.gromacs.org/pipermail/gmx-users/2004-July/011643.html )

I guess that gromacs on the PowerPC 970 with altivec loops will really
rock though, anybody tried already?

Greetings, Pim

--
Pim Schravendijk - PhD Student
Max Planck Institute for Polymer Research
http://www.mpip-mainz.mpg.de/~schraven/

> 
> Message: 1
> Date: Wed, 20 Oct 2004 10:06:24 +0200
> From: Stephan Frickenhaus <sfrickenhaus at awi-bremerhaven.de>
> Subject: [gmx-users] IBM p-series performance
> To: gmx-users at gromacs.org
> Message-ID: <41761C80.8020600 at awi-bremerhaven.de>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> 
> Hello!
> 
> I tested gromacs 3.2.1 (single prec)  on ibm-p655 power4 (1.1GHz) 
> (AIX5.1, xlc) systems on a single cpu.
> My testcase (3D box-solvated enzyme) shows around 600MFlops/s, which 
> appears low, compared to 1.85 GFlops on a Pentium4 prescott 
> 3GHz/800MHzFSB/PC3200 RAM, Intel8-compiler).
> 
> Any suggestions how to optimize? Is it possible to use the new 
> fma-version of fftw3.01?
> 
> Stephan
> 
> -- 
>        Dr. Stephan Frickenhaus
>   sfrickenhaus at awi-bremerhaven.de
> Alfred-Wegener-Institute for Polar and Marine Research
> Columbusstraße D-27568 Bremerhaven, Germany
> 0471-4831-1179 (Di, Do)
> 0421-218-4819 (Mo,Mi,Fr)



More information about the gromacs.org_gmx-users mailing list