[ESPResSo-users] different outcomes for CPU and GPU

[renata.tothova]

Hello, Espresso developers,

I wrote first email on 22 jun. I have encountered this problem:
I have run the simulation in periodic channel with fluid only. There were
only bottom and top wall and fluid flowed due to external force in x
direction in the channel. The values of velocity were differ, when I computed it with CPU or GPU.

Georg Rempfer wrote on 28 jul. “I ran you script and only got a difference of about 0.5%.”

In my case I compared values after 3000, 9000 and 12 000 integrate steps and there are relative deviations greater then 40%.
I have a box 100x50x30, so roughly 150000 fluid nodes and 148 nodes had relative deviation (CPU vs GPU) greater then 30%.

Have a nice day,

Renata Tothova

 

-- 
Renáta Tóthová
PhD student
Cell-in-fluid Research Group
KST FRI UNIZA