Think to try tNavigator software
They have free of charge multithread licensing
And the results of Eclipse and Tnavigator runs are almost the same (less than 1% by well diff)
In my company i have one permanent Tnavigator licence whick I use for setting up models ( blackoil, compositional, double porosity) and eclipse I rent of mothly basis just to make one final run and check everuthing is ok or not.
As for you question you will never guess how much processors you will need since the time you will spent for one run depends on many factors:
1. Number of active cells
2. Permiability variation (high values of perm will couse problems)
3. Way you initialise saturation in the model (cross flow of oil will distable you run)
4. Bubble point ( if the bubble point is near to you reservoir initial pressure , expect gas to be as free phase this will slow your model)
5. Aqifers (will slow your model)
6 Number of wells (more wells the longer is run)
7. Injectors (try to limit bhp of injector since the eclipse will extrapolate bph values to inject the water and this will couse the mat bal error since eclipse is first finding saturation)
8. Compositional run
9. Double porosity model run
But the main problem is the the errors in any cells since they will stop all you process since in order to finish step you need to solve matrix of saturation and pressure (temperature in composition run)
So as the rule of thumb ...
I have 24 core station but even this fact is not allowing me to have models more that 500 k of active cells
Normaly you will need at least 1 run for any well with history data to match it so you can easy calculate
field with 100 well - for 100 to 300 hours of setup (if you run is 1 hour) and 300-900 hours if you run is 3 hours.
So your strategy first should be targeting you time deadline.
Some one will say that we will lose heterogenity of the reservoir if we will coarse the grid too much
i always say the fact that the dimentions of cells of any grid is in the rang of 10s metres ( let say 50 X 50) and what we say is in the area of 250 m2 i will have the same porosity and permiability what is bullshit.
So instead of making 1 meter cell why I can not take thickness do 10 meters cell thickness?
The other question is the way you upscale….