AMBER Archive (2004)

Subject: AMBER: sander8 performance

From: Oliver Hucke (ohucke_at_u.washington.edu)
Date: Tue Jun 01 2004 - 12:53:46 CDT

Dear Amber users/developers,

I am using sander of amber8 compiled with ifort on dual processor nodes
of a linux cluster (Fedora 1.0). I am using the shmem device of mpich,
which gave me a speedup of a factor of ~1.5 compared to the ch_p4
device. I am working with Dual Xeon nodes running at 2.4 GHz, with 1 GB RAM.

With a sytem of ~50000 atoms, PME, cut=8.0, I get 75000 steps per day,
using two processors. This seems rather poor compared to the 117000
steps that were obtained in the hemoglobin benchmark with two Xeon
processors (2.0 GHz), with ifc7 compiled amber8, as I read on the amber
benchmarks page.

Do you have ideas on how to increase the performance?
What about using lam instead of mpich?
Should the hyperthreading capability rather be switched on or off?

I would like to use sander instead of pmemd because I want to compare
results from MM-PBSA with those from TI calculations - TI is not
possible with pmemd, right?

Thanks very much,
Oliver 

-- 
_______________________________________________________________

Oliver Hucke, Dr.
                               Health Sciences Building - K418C
University of Washington      1959 NE Pacific St.
Dept. of Biochemistry         phone: (206) 685 7046
Box 357742                    fax  : (206) 685 7002
Seattle, WA 98195-7742        email: ohucke_at_u.washington.edu
_______________________________________________________________

-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber_at_scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu