Closed AlexanderSinn closed 2 months ago
The problem is that most of the runtime is spent computing the fields at 4095*4095*112000
. When reducing the resolution by more than 10x the MG solver can stop converging, as the plasma frequency is not resolved properly. I don't think this is possible to run in ~2 min which would be acceptable for a CI test.
This PR adds inputs a transverse and a time step scaling benchmark. The transverse benchmark is tested in GPU CI with 1023^2 cells.
const
isconst
)