Re: NAMD on TACC (Darter)

From: Jim Phillips (jim_at_ks.uiuc.edu)
Date: Tue Dec 08 2015 - 22:41:19 CST

Sorry, the "-j 2" in the APRUNARGS enables hyperthreads, so there are 32
"cpus" per node. In my tests this was faster for smp runs (1 or 2
processes per node). Non-smp runs are faster with hyperthreads disabled.

Jim

On Tue, 8 Dec 2015, Chitrak Gupta wrote:

> Hi Jim,
>
> Thanks for update. However, I am still confused with the "+ppn 30 +pemap
> 0-14,16-30 +commap 15" line. Is it trying to use 32 nodes? But doesn't
> Darter have only 16?
>
>
> Regards,
> Chitrak.
>
> On Tue, Dec 8, 2015 at 11:15 PM, Jim Phillips <jim_at_ks.uiuc.edu> wrote:
>
>>
>> /nics/b/home/jphillip/NAMD_scripts/runbatch is the current script for
>> Darter. There is no runbatch_2.8 in that directory anymore.
>>
>> PPN is cores per node, PROCSPERNODE is processes per node.
>>
>> Jim
>>
>>
>> On Tue, 8 Dec 2015, Chitrak Gupta wrote:
>>
>> Dear NAMD developers,
>>>
>>> I am trying to run NAMD on TACC (Darter). I am looking at the runbatch
>>> script located at
>>>
>>> /nics/b/home/jphillip/NAMD_scripts/runbatch_2.8
>>>
>>>
>>> on Darter. I got this location from the following page
>>>
>>> http://www.ks.uiuc.edu/Research/namd/wiki/index.cgi?NamdAtNICS
>>>
>>>
>>> Could you verify whether this is updated for the current build on Darter?
>>> I
>>> am asking because looks like this one is trying to run on 32 nodes
>>> although
>>> Darter only has 16.
>>>
>>> Also, what is the difference between $PPN and $PROCSPERNODE variables in
>>> this code?
>>>
>>>
>>>
>>> Best regards,
>>> Chitrak.
>>>
>>>
>

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:22:18 CST