LAMMPS WWW Site - LAMMPS Documentation - LAMMPS Mailing List Archives
[lammps-users] The results from GPU and MPI LAMMPS were very different.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[lammps-users] The results from GPU and MPI LAMMPS were very different.


From: "shijy@...516..." <shijy@...516...>
Date: Wed, 27 Sep 2017 10:48:12 +0800

Hi all LAMMPS users,

I'm using LAMMPS (11Aug17) to simulate the interaction between dislocation and helium bubbles. I have perfomed the same case by both GPU (double precision)  and MPI versions of LAMMPS using the same input script and same cluster. But I got totally different reults for the stress-strain relationship. The stress-strain data for both GPU and MPI was attached. By the way, I also found that the difference is dependent on the characteristics of helium bubbles. When the content of helium atoms in helium bubbles is low, the difference is small. Conversely, the result is very different. 

PS: After I checked the mail list, I found the double precision calculation is very useful to remedy the difference between GPU and MPI in some case. But in my case, I already used the double precission to compile the GPU version, it still didn't work.

Thanks so much for all your comments and advice.
 

Best wish for you!
Daniel Shi

Attachment: Stress_GPU_Double.txt
Description: Binary data

Attachment: Stress_MPI.txt
Description: Binary data