Re: AMBER: amber8 performance problem on Itanium II's

From: Andreas Svrcek-Seiler <>
Date: Fri, 17 Sep 2004 10:52:53 +0200 (CEST)

> I compiled amber8 (both the serial and REM versions) on the front node
> of an Itanium II (1.3 Ghz) cluster (SDSC). The intel compilers were used
> and the MKL libraries included. However, in comparison with amber7,
> amber8 runs about 4 times slower (~73 ps/day for a 3700 atom system using
> GB=1). The nonbonded cut-off = 12.0 A and used ntt=1 (just as in the
> amber7 runs).

....It's been quite some time since I dug around in egb.f from amber8,
so this is a semi-educated guess:-)
In amber8, there are -technically- two cutoffs, once for the
pairwise (inverse)
effective born radii contributions ("rgbmax", defaults to 25 A) and the
old "cut", for the usual non-bonded interactions. While cut=12 A might be
a bit small, it should be ok for rgbmax (even 9 A seems ok, and even
for nucleic acids). You simulation should be much faster by setting
e.g. rgbmax=12 A and cut= 15 A. See the manual, around page 99.
With the above settings, your MD should run faster than with amber7
(I hope).

I hope that helps,
good luck

            ( O O )
               o Wolfgang Andreas Svrcek-Seiler
               o (godzilla)
       .oooO Tel.:01-4277-52733
       ( ) Oooo.
-------\ (----( )--------------------------------------------------------
         \_) ) /

The AMBER Mail Reflector
To post, send mail to
To unsubscribe, send "unsubscribe amber" to
Received on Fri Sep 17 2004 - 10:53:00 PDT
Custom Search