port to GPU
It may be possible to port part of the code on GPU for configurations that are not using the energy solver.
Here is a typical config file config4.info
for a simulation I am running on Licallo. With the requirement in subm4.sh I have about 400 steps in 24 hours (on 320 cores)
The resolution is moderate. I have a similar test on double resolution in each directions (8 times more grid cells) running on Occigen. With the requirement of 10 nodes which corresponds there to 240 cores (20 MPI processes each on 12 cores using openmp) I have about 30 steps in 24 hours. It is not the right way to compare (different machine, less core for the high res. case ...), but it gives an idea of the challenge of increasing the resolution on simulations requiring long integration times