From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Wed Jun 14 2006 - 18:24:28 CDT
The reported wallclock time for NAMD 2.6b1 and earlier is confused by
machines with variable clockspeeds. (It checks the CPU speed at startup,
and then multiplies by the number of cycles to get walltime cheaply.)
This is fixed in newer versions of Charm++ (and hence NAMD). A workaround
is to have some kind of load on the machine when starting the benchmark,
or to just multiply by the ratio of the resting and loaded clockspeeds.
On Wed, 14 Jun 2006, Fam Eklund wrote:
> I have resently upgraded my computer to an athlon64 +3500 1GB ram, using an ASUS A8n5x motherboard. And I am trying to run NAMD in linux mode. I have installed the new SUSE 10.1 (for 64 bit computers).
> I have both tried to run NAMD from the 64bit binary that one can download from their homepage, as well as from the compiled one from source, and get almost the same result.
> The program starts and runs pretty smoothly however when I check the benchmark for the run I get the following :
> CPU TIME 0.87s/step WALL time 1.99s/step
> Why is the total time more then twice the cpu time, is the computer .
> (I have also tried to run the same simulation on my windows partition using the windows binary (32bit?) from the NAMD homepage, and run it on windows XP home edition. I get the following results CPU TIME is 1.3 s/step and wall time is roughly the same.)
> Shouldn't the linux version be faster then the windows version ???...
> If someone has encountered the same problem, or has any idea about what could slow the process down please let me know. Because if we are going to expand the system and connect more computers we want to know if we should continue with the athlon64 type or if it is this slow try something else.
> Robert Eklund PhD Organic chemistry Stockholm University.
This archive was generated by hypermail 2.1.6 : Wed Feb 29 2012 - 15:43:42 CST