[eliwardian]

Hi, I like to know what is the best core to ram ratio. For example, 1 core to 1GB - 10GB. Thanks in advance.

[flotus1]

In my opinion, the relevance of this metric is debatable at best.

There is only one real constraint: the total amount of memory required to fit your model.
If the simulation requires 100GB of RAM, then the machine needs 100GB (typically populated as 128GB) of RAM.
You can run it regardless of how many cores the machine has. Sure, having more cores will tend to solve the model faster. But the real bottleneck here will be memory bandwidth, as proven time and time again in the sticky thread in the hardware sub-forum. This means that at a certain point, CPUs with higher core counts will not solve the model any faster.
An extreme example of this is the TR 3960X CPU added recently. Here are the numbers:

# cores Wall time (s):
------------------------
1 550.49
2 299.15
4 161.65
6 120.55
8 101.56
12 99.13
16 93.74
20 93.71
24 93.65

As you can see, scaling falls off after 8 cores. Different CPUs with the same architecture achieve much better scaling (and overall performance) due to having more memory bandwidth available.
Switching from the 24-core Threadripper CPU to its 32-core variant (or even 64 cores later down the line) would change the "cores/GB" ratio, but do absolutely nothing for the solver performance.

In conclusion: I would separate the decision for system memory and CPU.
Get the CPU(s) that perform best within your given budget. And make sure you have enough memory to fit your largest simulations.

Edit: this should be moved to OpenFOAM or hardware sub-forum