optimum number of processor

I am new to MRCC programme. I am doing few test calculations using ccsdt and ccsdt(q) methods. I am using Oenmp parallelization to use the number of cores present on my workstation. My system has 48 cores and 100 GB Ram. I realized till 8-10 processor speed of job increase but beyond that there is no effect of processors on the speed of job. In fact when I make number of processor 20-30 speed of jobs decreases. Can someone help me regarding that. Is problem is with Openmp or CCSDT methods can not use more processors??????????

The parallelization efficiency depends on the size of the job, and it is better for bigger ones. Nevertheless, the parallelization efficiency of iterative CC methods, such as CCSDT, is not so good, and probably it is not worthwhile running such jobs on more than 10 cores. The parallelization efficiency of perturbative corrections, such as (Q), is much better, and more cores can be used for such methods.

It probably depends on many factors. My impression is that fast scratch disks (SSDs or even ramdisk) are essential for good overall performance.
Additionally the architecture of your machine might be important. A high number of NUMA nodes might degrade your performance in OpenMP cases. At least this was my impression when comparing timings from machines with 2 vs. 8 NUMA nodes.