[gmx-users] Scaling on IBM/SP2 Power3/4

[Anton Feenstra]

Hi All,


A week or so ago I reported on bad scaling on an IBM/SP2 Pwr3 machine.
This has been partly resolved by running 'grompp -shuffle -sort':

For the DPPC benchmark system I get the following behavior:

        'normal'        'shuff+sort'	Power4
#CPU   ps/h scale       ps/h scale	ps/h scale
 1      2.0  1.00        2.0  1.00	 3.7  1.00
 2      3.3  0.81        4.2  1.03	
 3      4.5  0.74        6.4  1.04	
 4      5.3  0.66        8.5  1.04	18.3  1.23
 6      8.2  0.68       10.5  0.86	
 8                      16.3  1.00	
10                      20.3  1.00	
16                      31.3  0.96	
24     			42.2  0.86	
32     			50.7  0.78	

E.g. for 4 CPU's scaling goes from 67% without shuffle/sort to 104%
with! Even for 32 CPU's (two nodes with 16 CPU's each), scaling is
still near to 80%! The Power4 Nodes are really outrageous: scaling
is up to 125% (!) for 4 nodes relative to 1. That means, that four
nodes give you 5 (five!) times the performance of 1 (one!).

You should of course keep in mind that the DPPC benchmark system is
a really optimal system for scaling purposes: 1024 lipids, lots (!)
of water and no protein. For my own project I want to run a simple
protein + ligand + water system (40000 atoms), for which even 4 CPU's 
only scale 75% (56% on 8 CPU's), and -shuffle/-sort does not improve...
Haven't tried that on the Power4 nodes, though.

On the assumption that many Gromacs users would be interested in 
these kinds of benchmark results, I'll keep you posted!


-- 
Groetjes,

Anton
 ________ ___________________________________________________________
|        | Anton Feenstra                                            |
| .      | Dept. of Pharmacochemistry - Vrije Universiteit Amsterdam |
| |----  | De Boelelaan 1083 - 1081 HV Amsterdam - The Netherlands   |
| |----  | Tel: +31 20 44 47608 - Fax: +31 20 44 47610               |
| ' __   | Feenstra at chem.vu.nl - http://www.chem.vu.nl/afdelingen/FAR|
|  /  \  |-----------------------------------------------------------|
| (    ) | Dept. of Biophysical Chemistry - University of Groningen  |
|  \__/  | Nijenborgh 4 - 9747 AG Groningen - The Netherlands        |
|   __   | Tel +31 50 363 4327 - Fax +31 50 363 4800                 |
|  /  \  | K.A.Feenstra at chem.rug.nl - http://md.chem.rug.nl/~anton   |
| (    ) |-----------------------------------------------------------|
|  \__/  | "I'm Only Faking When I Get It Right" (Soundgarden)       |
|________|___________________________________________________________|