CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   FLUENT (http://www.cfd-online.com/Forums/fluent/)
-   -   3d & 3ddp (http://www.cfd-online.com/Forums/fluent/99413-3d-3ddp.html)

pazhouhesh.m.j April 3, 2012 04:19

3d & 3ddp
 
Dears,

Would you please,tell me what is the difference between 3d & 3ddp in fluent?
The results i got from 3d & 3ddp are different,I mean the difference between them are large.So which results is reliable? or are the 3d results wrong?

Sincerely,
Pazhouhesh

kitrax April 3, 2012 04:58

Hi,

3d results are single precision floating point numbers and 3ddp are double precision floating point numbers. You normally use double precision in a Fluent solver.

It works like this:
In single precision: This is a binary format that occupies 32 bits, has a precision of roughly 7 decimal places.

In double precision: occupies 64 bits, has a precision of 16 decimal places.

The main idea you have to take from this is that over a series of iterations, the errors in single precision (3d) could be significant, hence your 3ddp is the solution you should use :)

Hope this helps, if it does, increase my rep power B)

K

pazhouhesh.m.j April 3, 2012 05:04

Dear,

There is not any error in my problem (both 3d & 3ddp),only a big difference is in the results.So does it mean that ,3d of fluent does not work properly?

Sincerely,
Pazhouhesh

kitrax April 3, 2012 05:34

Well depending on the number of iterations, I would think 3d is not suitable, if you have a very coarse mesh, I think 3d would be fine. Though for a fine mesh, the error of 3d is carried forward through each iteration therefore leading to a difference between 3d and 3ddp.

What you can do is run your system with a few iterations with each precision and compare the 2 numbers, though I would say if you have already done 3ddp just stick with it :)

Btw what are you studying?

LuckyTran April 3, 2012 10:29

Quote:

Originally Posted by kitrax (Post 352864)
Well depending on the number of iterations, I would think 3d is not suitable, if you have a very coarse mesh, I think 3d would be fine. Though for a fine mesh, the error of 3d is carried forward through each iteration therefore leading to a difference between 3d and 3ddp.

What you can do is run your system with a few iterations with each precision and compare the 2 numbers, though I would say if you have already done 3ddp just stick with it :)

Btw what are you studying?

Just use double precision. There is no reason not to, it is standard nowadays. The difference between single precision and double precision should be hardly noticeable, but just use double precision. It is outright superior.

If the solution is converged, the number of iterations should not affect the accuracy of the single precision vs double precision in the sense of the error being carried forward (unless the simulation is transient). For steady state solver there is no difference, except in the machine precision. i.e. we define convergence based on the solution value and not on the number of iterations that have passed.

ashrawage November 11, 2012 13:43

About double precision in piecewise polynomials
 
Hello there,
I was just hoping you guys would know about how to increase the decimal places in the piecewise polynomial box where you enter the coefficients for your nth order equation for predicting material properties.
In my case i am using Double precision but for some reason the coeff. are getting rounded off and believe it or not; that reduction in the decimal places messes up my property prediction big time! (i plotted curves using MATLAB and saw the difference)
Please let me know how can I increase the decimal places in the piecewise polynomial boxes

thank you
Abhijeet



ashrawage November 11, 2012 13:44

Images
 
2 Attachment(s)
I have attached the matlab vs fluent images for specific heat




All times are GMT -4. The time now is 11:19.