Dear LAMMPS users,
We just released a new patch release of LAMMPS with internal improvements and several new packages. You can see an overview of the included changes here:
https://github.com/lammps/lammps/releases/tag/patch_14May2021
One important aspect is that several developers have contributed bug fixes and improvements to the GPU package. So if you are using LAMMPS version 2 Feb 2021 or later with the GPU package, you probably want to upgrade to benefit from those bugfixes (if you are using an older version of LAMMPS you may want to upgrade for the better support for AMD and Intel GPUs and better GPU acceleration on those).
There are still a few known and possibly some more unknown issues in the improved GPU package, but to track them down we need your help. If you have a configuration that is not listed at https://github.com/lammps/lammps/issues/2747, please try to compile LAMMPS using CMake with the GPU package include and unit testing enabled, and run those unit tests with “ctest -V”. See https://docs.lammps.org/Build_development.html#code-coverage-and-unit-testing for more info on the unit tests.
Please report your results by replying to this message or adding a comment to the issue on GitHub.
As a reminder, the GPU package can be compiled with three (3) different back ends using either Nvidia’s CUDA, or AMD’s ROCm/HIP, or OpenCL (no toolkit or library required to compile, LAMMPS will use the official OpenCL driver loader library automatically, except on MacOS), and the OpenCL version is tested to run on GPUs from Nvidia, AMD, and Intel (the ones integrated in the CPUs).
Thanks in advance for using LAMMPS and participating in making it an even better software than it already is. 
The LAMMPS Developers.