[gmx-users] --enable-threads fails.
lindahl at cbr.su.se
Wed Apr 11 08:29:53 CEST 2007
The prefetching hardware is pretty good at hiding latencies, but
obviously you'll get better performance when everything fits in cache.
However, the largest data structures by far are the neighborlists,
and their _total_ size for the simulation is independent of whether
you are using multiple processes or multithreading.
On Apr 10, 2007, at 10:46 PM, vipin sachdeva wrote:
> Thanks. Do you see L2 cache overflowing effects by putting multiple
> MPI processes on the same chip with the recent multi-core processors
> (Intel Woodcrest for instance)?
> On Tue, 10 Apr 2007, Erik Lindahl wrote:
>> On Apr 10, 2007, at 10:15 PM, vipin sachdeva wrote:
>>> This thread on Feb 8 2006 mentions that threads are not yet
>>> Is the threads still not functional considering I still get the
>>> same error ?
>> We are only using threads for experimental development of the
>> nonbonded interaction loop. The multi-level-cache-locality
>> features on modern CPUs (even NUMA on Opterons) means that you
>> normally get better performance by partitioning the problem into
>> separate MPI processes, where each process resides on a separate
>> gmx-users mailing list gmx-users at gromacs.org
>> Please search the archive at http://www.gromacs.org/search before
>> Please don't post (un)subscribe requests to the list. Use thewww
>> interface or send it to gmx-users-request at gromacs.org.
>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
> gmx-users mailing list gmx-users at gromacs.org
> Please search the archive at http://www.gromacs.org/search before
> Please don't post (un)subscribe requests to the list. Use the www
> interface or send it to gmx-users-request at gromacs.org.
> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
More information about the gromacs.org_gmx-users