Precision of performance

Dear lammps users,

I have questions about the precision of LAMMPS.
For example, I downloaded an executable LAMMPS for Linux on my PC.
Therefore, I use the command “lmp_stable”.

When I use LAMMPS with GPU, the precision is shown as “Mixed Precision”.
But I wonder that the precision of the simulation is limited by single-precision due to the single-precision calculation.
I want a comment for this question.

Also, I build LAMMPS on the other PC by cmake.
The way to build is simply because I make a directory and a command “cmake …/cmake”.
But I don’t specify the precision in the above build.
In this case, which is the calculation performed by single- or double-precision?

Best regards,