[gmx-users] OPLS C-terminal Capping of Glu residue a bug?

Erik Lindahl lindahl at stanford.edu
Sun Aug 24 05:55:01 CEST 2003


Short answer: Just compile in 32-bit mode and the Opteron performance 
will be excellent (using SSE assembly loops).

Long answer: The current gromacs version doesn't have any 64-bit 
opteron assembly loops, but I have very strong reasons to believe 
future versions will be (read: I have already written the x86-64 
assembly loops :-). In practice they won't be any faster than the 
32-bit loops, and in some cases it might actually be slower due to the 
64-bit pointers. This doesn't mean the 64-bit mode is bad, it is just 
because the AMD chip is really excellent at running the 32-bit assembly 
code! (and it is in agreement with other x86-64 benchmarks). Future 
generations of x86-64 might be more focused on 64-bit performance 
though, so it might matter in a generation or two, and they 64-bit 
loops will make it easier to get good performance on huge systems 
(millions or billions of particles).

Just make sure you use 32-bit mode, and that the log file says 
something about SSE (so you are not running 64-bit without assembly 



> Hi, Erik
> I am just wondering whether Gromacs performance on Opteron has been
> optimized.
> 1. yes, then do I need to use a special option for compilation.
> 2. no, I hope it happen before this Christmas.
> If I do some BM on Opteron with Gromacs 3.1.5_pre1, it seems to me the 
> data
> is useful for just comparison(before/after optimizaton), not reliable 
> to
> evaluate Opteron as decision making data.
> Thanks,
> taeho
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.

More information about the gromacs.org_gmx-users mailing list