Gaussian: Benchmark of G16 running on Intel Xeon Gold 6148

In the early 2018, Gaussian released G16 B.01 after releasing revision A.03 on 2017. I did a benchmark of this latest version of G16 in density functional theory (DFT) calculation in comparison with the previous version, G09 revision E.01. As I read this post on http://computational-chemistry.com. It really inspires me to do this benchmark, which might be useful for consideration in choosing the Gaussian runtime.

To be fair, I adopted the Gaussian input file from previous benchmark. I separately calculated geometry optimization and frequency of Vomilenine using DFT method.

Computational details

Linux OS: Red Hat Enterprise Linux Server release 7.3 (Maipo)
CPUs model: Intel(R) Xeon(R) Gold 6148 CPU @ 2.40GHz (Total physical/logical cores = 20/40 cores)
Memory: 8 GB (for all calculations)
Software: G09 E.01 and G16 B.01
Parallel method: Shared memory (OpenMP method)

Input file

Input file can be obtained from https://pastebin.com/ja4s7e97

All calculations employed memory of 8GB. For G16 and G09, I leaved their calculation running with the default parameter. But, G09-tight, I run G09 with the keywords int=grid=untrafine and scf=tight, which increases the number of grid and enhanced integration accuracy from 10^-10 to 10^-12.

Benchmark results

Benchmark-G16b01-G09e1-Xeon

Concluding remarks

My benchmark results show that G16 is running slower than G09 becuase default setting of G16 uses SCF convergence criteria and number of grid point higher than G09.
To force G09 to follow the default setting in G16, int=grid=ultrafine and scf=tight should be used. However, G09+these external keyword is still faster than G16.
New features in G16 is improving the performance of some method. The efficiency of most used method and SCF algorithm in G09 is, compared with G16, still powerful.
G16 does not significantly improve its speed in calculation, particularly DFT, you maybe can use G09 for your further calculation.
However, most of new features in G16 are state-of-the-art method that computational chemist has in present.

Rangsiman Ketkaew