[gmx-users] Athlon Instabilities

[Erik Lindahl]

Hi Stefan,

I too suspected a bug in the AMD SSE implementation at first - that's 
why I wrote the cpu heating program on the contrib page of 
www.gromacs.org. However, after running it for weeks on two of our 
machines it didn't find any errors at all, so I guess the SSE part of 
the Athlon CPU is OK, after all.

On Thursday, May 22, 2003, at 08:16  AM, Stefano Piana wrote:
>
> Now the question is:
> I am not too familiar with the gromacs code; could it be that even if 
> I configure with the --enable-fortran --disable-x86-asmc flags, SSE 
> instructions are still used by the program?

No, the GNU compiler at least does not generate any SSE stuff - the 
Intel compiler might, though.

> Is it normal that gromacs produces slightly different results (0.1-1%) 
> already after 100 steps of MD when the same calculations is run twice?

Computer programs are deterministic, so if you use the same tpr file 
and run on the same machine the results should be identical. It is 
normal to get very slight differences between different CPU (even Intel 
and AMD) since they might perform the instructions in a slightly 
different order (dynamic instruction scheduling).

> Could it be simply an heating problem that comes out more often when 
> sse instructions are called?

Something like that (running tight loops with both SSE and integer 
stuff heats the CPU more), but since my test program didn't find it on 
our machines it is probably dependend on the memory controller or 
something too. I'm still checking this, but I'm running out of ideas 
(everything works perfectly on Pentium/Xeon boxes).


Cheers,

Erik