4:53:40 AM PST - Wed, Feb 13th 2019 |
|
Hello again, please excuse the late reply. I had troubles running the input and did some extensive testing on different molecules and hardware.
This specific error seems to appear for a range of SnGd molecules and the input file works for smaller molecules (which previously produced the same problematic output). However, starting with 12 atoms I was not able to run the calculations and with that neither the Sn14Gd input. I tested the input on different HPC, in each case the slurm system reports that the RAM demand is too high. I tested this for different hardware:
-1 node with 250 GB RAM per node
-1-5 nodes with 26 GB per node
I additionally tried to add the direct keyword in front of the grid keyword, but in each case the error appears unchanged. I also tried to measure the RAM costs with the slurm command: sacct -j SLURMID--format=jobid,elapsed,ncpus,ntasks,state,MaxVMSize,AveVMSize
The command shows that the jobs consume far lower RAM than provided, but still the jobs crash with RAM error.
Is this really a RAM issue or something else appearing as RAM error? How much RAM should these jobs actually consume? Is there away to reduce the required RAM?
Thank you in advance!
|