[gmx-users] Strange self-coefficient data with "-update gpu" option

[qing shao]

Hey,
We ran a quick simulation of a box of 884 spce water model using Gromacs
2020 on a workstation with an RTX 2080 card.
The water box was generated using gmx solvate command.
The whole simulation is 10 ns NPT + 10 ns NVT.
The self-diffusion coefficient is   D[        OW] = 23.5350 (+/- 11.1866)
(1e-5 cm^2/s) if the MD simulation uses the "-update gpu" option
The value turns to normal as D[        OW] = 2.5914 (+/- 0.0069) (1e-5
cm^2/s) without the "-update gpu" option.

Just wondering if anyone observed similar things.

Best,

Qing