Hi, there,
 
We were running ADF 2002.03 at our Athlon Linux clusters that have dual
processors. The calculation is very fast and scales impressively well. The
clock is 1200 MHz. Here is the timing information for a molecule with U, C,
O, H atoms:
 
For 1-processor (all time in second):
 
Dirac:
 
 U: Total Used : CPU= 1.39 System= 0.04 Elapsed=
1.44
 C: Total Used : CPU= 0.15 System= 0.02 Elapsed=
0.18
 O: Total Used : CPU= 0.13 System= 0.04 Elapsed=
0.18
 H: Total Used : CPU= 0.10 System= 0.01 Elapsed=
0.11
 
Create:

U: Total Used : CPU= 1.90 System= 0.09 Elapsed=
2.08
C: Total Used : CPU= 0.55 System= 0.03 Elapsed=
0.58
O: Total Used : CPU= 0.58 System= 0.03 Elapsed=
0.61
H: Total Used : CPU= 0.38 System= 0.06 Elapsed=
0.43
 
The ADF run for a U-C-O-H containing molecule (146 SFOs, C2v symmetry,
opt/zora/tz2p):

 Total Used : CPU= 561.18 System= 6.04 Elapsed=
569.54

 
The ADF run on the molecule with 6-processors:
 Total Used : CPU= 103.34 System= 23.14 Elapsed=
133.99
 
Jun Li
 

-----Original Message-----
From: Tris Youngs [mailto:t.g.a.youngs_at_reading.ac.uk]
Sent: Friday, January 10, 2003 6:34 AM
To: adf-list_at_scm.com
Subject: ADF-LIST: ADF runs VERY slow on dual Athlon box.

Hi.
 
We have a small cluster here, and have installed and successfully run ADF on
single processor 1 Ghz Athlon machines.
 
However, we have some dual Athlon boxes (1900+ MP processors) and ADF is
unusable! The boxes are not set up to run as a massive parallel machine, nor
am I trying to run ADF on more than a single processor (at this time).
Running a single instance of ADF on the dual box, with no other big
processes running, is painfully slow - a simple DIRAC calculation at the
start can take up to a minute for a light element!
 
Obviously there is something seriously wrong with the setup of the dual box
(other programs, such as Gaussian, run without problem).
 
Can anyone shed any light on this?
 
Cheers,
Tris.
 
Department of Chemistry
University of Reading
Whiteknights
Reading
England
Received on 2003-01-10 18:20:57