Fastest way of running GridSearchCV with multiple CPUs

I need to perform GridSearchCV with 4 different classifiers. I have at my disposal 128 CPUs which i can use in different sessions (2 sessions with 64CPus, 4 sessions with 32 CPUs, etc).

I'd like to ask if there's any way of knowing which could be faster:

• Run 1 session w/ 128CPus 4 times;
• Run 2 sessions w/ 64CPUs in parallell 2 times;
• Run 4 sessions w/ 32CPUs in parallel 1 time.

from Fastest way of running GridSearchCV with multiple CPUs