<div dir="ltr">Dear Samuel, thank you for your reply. about the memory: Yes, it is possible to run the code on more nodes and the supercomputer has different partitions which belong to several institutes and our partition contains 180 nodes with 12 cores each. But for my project, I have reserved / allocated a certain amount of CPU times and therefore I would like to check first small simulations (how many steps per day and how efficient the simulation was run) before I start large simulations. I am not sure how to write a SLURM code to use one rank per node. Should the code look like or should I define OMP_NUM_THREADS and ntasks differently ? : ------------------------------------------------------ #SBATCH --nodes=16 #SBATCH --tasks-per-node=12 #SBATCH --cpus-per-task=1 #SBATCH --mem-per-cpu=3000M echo Starting Program module load cp2k module load bullxmpi export OMP_NUM_THREADS=4 srun --ntasks 192 cp2k.popt input.inp echo Finished Program ------------------------------------------------------- linear scaling: Here is the newly updated input file. Should the part under the section &SCF be different (make it shorter) or does the dioganlization process in SCF influence the linear scaling method negatively? I mean, if the linear scaling SCF is running then the (conventional) SCF works simultaneously, doesn't it? Otherwise I can delete most entries under this section. What do you think? Kind regards, Phil </div>