On Fri, May 19, 2017, kurisaki wrote:
>
> My umberlla sampling simulation with Amber 16 in GTX1080
> ofter quit anormalously.
>
> Example in the below, shows overflow of Fortran format
> Between 130000 and 135000.
What you report (generally) is not all that unusal. I've had GPU simulations
that suddenly fail after hundreds of thousands of steps, and for reasons
I am unable to figure out. My hunch (just that) is that somehow a large
force is overflowing the fixed-point representation used for reductions.
You could see what happens with a CPU simulation, starting at a point right
before the GPU explosion.
I recognize that this not much help.....dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun May 21 2017 - 17:00:03 PDT
