Monitor usage:
Exceeding a certain core count can actually performance due to increased thread management overhead and memory contention. A systematic scaling study for each unique system and method is essential; wasting cores does not speed up the calculation. gaussian 16 linux
setenv g16root /opt/gaussian setenv GAUSS_SCRDIR /tmp/g16_scratch source $g16root/g16/bsd/g16.login Monitor usage: Exceeding a certain core count can
export g16root=/opt/gaussian16 export GAUSS_SCRDIR=/scratch/$USER/gaussian # fast local disk source $g16root/g16/bsd/g16.profile Benchmarks have shown that for large molecular systems
Gaussian 16 can leverage NVIDIA GPUs (K40, K80, P100, V100, A100) to accelerate certain calculations, such as DFT and HF. Benchmarks have shown that for large molecular systems (~1 million atoms), GPU acceleration can provide up to a compared to using CPUs alone. To enable GPU support, the job script must load the correct module that includes GPU capabilities and may require specific runtime directives.
Gaussian 16, Revision A.03, M. J. Frisch et al., Gaussian, Inc., Wallingford CT, 2016.
Also verify that rsh or ssh works without password: