Hi Albert,
>That's since to be nice. Does anybody also consider improve the CPU
>performance of Amber? I can only get max. 10 ns/day for normal 50,000
>protein/water system with cutoff=0.8 while I can easily get upto 30-40
>ns/day with another two MD tools with cutoff 1.0 under Amber99-ILDN FF.
>That's really headache for the CPU poor performance of Amber. Hopefully
>it can also be improved in the near future.
I suspect you are not doing an apples to apples comparison in terms of CPU
performance. There is a big (but sometimes subtle, or not well described)
difference between improving performance and increasing the number of
approximations used. If you are comparing to Gromacs then for a fair
comparison you should use the double precision version of the code, only
shake hydrogen bond lengths, use a 2 fs time step, make sure you are using
PME with the same grid spacing, ewald coefficient and interpolation order,
make sure you are doing the PME on every time step and use either NVE or
the same thermostat if supported in both codes. Otherwise you are not
comparing the same thing.
If you want to compare to NAMD please see the following page which
explains how to run the AMBER force fields correctly in NAMD:
http://ambermd.org/namd/namd_amber.html
Note the benchmark on that page which is for a reasonably good apples to
apples comparison. The same caveats apply, use the same level of,
conservative, approximations in both codes.
Now as to whether one should make extensive approximations to the
underlying mathematical model in search of greater performance that is a
very big can of worms and a discussion for another day, ideally over a
reasonably large number of pints of beer.
All the best
Ross
/\
\/
|\oss Walker
---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Sep 14 2012 - 09:30:02 PDT
