Problem with tersoff/cuda


can you give me some more info with regards to your run?
I run your input and got the result as attached. With double precision I got no outliers, with single and mixed precision I got one each. While this doesnt look as bad as your output its still strange. That there is no spike in the kinetic energy suggests that it probably is some error in the energy accumulation or so - something which would only affect the reported energy not the forces. Otherwise I'd expect some lost atom or so. I am going to look at it closer tomorrow.

But could you copy and paste the output before the run actually began.
Also can you provide CUDA version and Drivers version? (Try nvcc --version, and nvidia-smi -a | grep Driver).

Thanks for your help

-------- Original-Nachricht --------