what PC server or workstation can run so big model?

[wisdomxuhl]

hi, friends,
I'm preparing for a dynamic simulation, but firstly, no computer.
my dynamic model:
0.8-1.0 millions active grid blocks,
number of wells: 150-200
type of model: 1. blackoil; 2. component model;

my question is how many parallel licenses I should use for this model ,
in addtion, what PC server or workstation can run so big model?
pls give some advice, thanks in advance.

[herbertharturo]

Depend of which simulator you will use.

[temr]

first you need to think about 64 bit version of Simulator and operating system
the culculation i simple each active cell will requiere 3k of RAM so in your example you will need computer with 64 bit operating system with RAM for eclipse process of 3 GB , to make you able to run other windows task i will advise you RAM of 16 GB minimum
Then test your model in single mode since parrelel is not always will mean that you will defently short time for run since if you will have problems in one of multiple threads that will slow all run.
Fro, my expirience the more comfatable number of active cells for blackoil is maximum to 1 m
for compositionla model (8 comp) - 400 - 450 thousands active cells

[coby]

Even a 2010 PC can handle a million cells model, it just takes more time than the 2014 PC.

For 2014, it is not uncommon to have a 128G multi-core PC for simulation engineer.

[coby]

Even a 2010 PC can handle a million cells model, it just takes more time than the 2014 PC.

For 2014, it is not uncommon to have a 128G multi-core PC for simulation engineer.

[temr]

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….

[coby]

Nowadays, we hardly run full field study. The FF basically for the show. When needed, we farm the run out. The one time fee may seem high but still make more economic sense.