[gmx-developers] threads are now ON by default

XAvier Periole x.periole at rug.nl
Fri Feb 12 18:46:17 CET 2010


This file that was attached did not go through I guess,
so I send my comments again :))

On Feb 12, 2010, at 6:14 PM, David van der Spoel wrote:

> On 2/12/10 5:32 PM, Sander Pronk wrote:
>> Now that the last issues have been resolved with the threading  
>> code, thread-based parallelization has been turned on by default.  
>> To disable all the threading code, use
>>
>> --disable-threads.
>>
>> in configure, or turn the option GMX_THREADS off with ccmake.
>>
>> Running mdrun with just one thread (the default) is almost exactly  
>> the same as running it without threading code: the only thing  
>> that's different is that the few remaining global variables are  
>> protected by mutexes.
>>
>> Performance-wise mdrun runs very slightly faster with threads than  
>> with OpenMPI when Nthreads<=Ncores (and there is no other processes  
>> on the computer). When Nthreads>Ncores (or other processes are  
>> running), the thread code is much faster than OpenMPI, but the  
>> total runtime is still smaller than when Nthreads==Ncores.
>>
>> If there's any problems in getting things running, or with  
>> performance, I'd very much like to hear about it.
>>
>> Sander
>>
>>
> Great work Sander!
>
> What about the rumours that enabling hyperthreading gives you  
> superscaling beyond Ncores?
my experience on poweer-PCs is that using hyperthreading indeed gives  
you
back some of the lost performance due to unperfect scaling. With few  
cpus you get
superscaling :))

In France they also have a power-PC supercomputer and use the  
hyperthreading
as default ...

I attach a file showing some tests I made some time ago.
(Huygens is the power-PC thing


>
> -- David van der Spoel, Ph.D., Professor of Biology
> Dept. of Cell & Molec. Biol., Uppsala University.
> Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205. Fax: +4618511755.
> spoel at xray.bmc.uu.se	spoel at gromacs.org   http://folding.bmc.uu.se
> -- gmx-developers mailing list
> gmx-developers at gromacs.org
> http://lists.gromacs.org/mailman/listinfo/gmx-developers
> Please don't post (un)subscribe requests to the list. Use the www  
> interface or send it to gmx-developers-request at gromacs.org.

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20100212/e2021457/attachment.html>


More information about the gromacs.org_gmx-developers mailing list