|
[Sponsors] |
December 5, 2000, 12:20 |
Memory needed for big case
|
#1 |
Guest
Posts: n/a
|
Hello CFD guys !
I need to define the hardware setup necessary for some problem I am solving now, but at the moment I have not enough memory (RAM) on my computer. I would like to know if someone has experience with big case (> 400 000 cells) calculations using DO radiation model (+ turbulence + multiple species). If you have something like that I would like to hear the hardware you use (memory and CPU) and the time necessary to complete simulations. Thanks for your answers. |
|
December 7, 2000, 11:50 |
Re: Memory needed for big case
|
#2 |
Guest
Posts: n/a
|
Hi, for 400.000 cells, turbulence (2Eq.), Energy and 3 species you should need about 400 to 440 Mbyte RAM. For the radiation I estimate (!) additional 40 MB.
|
|
December 8, 2000, 04:37 |
Re: Memory needed for big case
|
#3 |
Guest
Posts: n/a
|
Thanks for your answer Volker.
But I must say that I do not agree with your estimation of aditional memory for radiation. From my experience after some testing, memory needed for the use of radiation modeling + species ranges from +40% to +70% depending on the mesh size. Had someone else experiences with such cases and confirm my estimation??? |
|
December 8, 2000, 08:03 |
Re: Memory needed for big case
|
#4 |
Guest
Posts: n/a
|
Hello Tschumperle,
meanwhile I had a look on an old case file with k-eps, energy, 1-species a) with and b) without DO-Radiation: The case had 896677 cells. The RAM usage with Fluent 5.4.8 was for a) 852548 MB b) 795568 MB Hence I think that an estimation of about 10% from the number of cells = additionnal RAM for the DO -Modell is ok. As far as I know the DO gives one more governing equation and 10 % for 1 eq. is typical for what I observed in the past. But maybe you changed the std. parameters of the model? Isn't there a switch wich controls the facette resolution? |
|
December 8, 2000, 13:04 |
Re: Memory needed for big case
|
#5 |
Guest
Posts: n/a
|
I checked the Fluent 5 manual and my understanding is that for ntheta x nphi discretization per octant, there are 8*ntheta*nphi additional scalar equations being solved, and the half in 2D. And that is of course in the case of a grey model (no bands). You can control this octant discretisation and it is advised to choose it 3x3 in the presence of transparent bodies. So I hope that 10% for an additional equation is not a general rule, because you see that the memory jump would be a whooping 240% for a typical case.
|
|
December 11, 2000, 04:06 |
Re: Memory needed for big case
|
#6 |
Guest
Posts: n/a
|
1. a correction to my last message: The RAM requirement was in kbyte
a) 852.548 MB or 852548 kbyte b) 795.568 MB or 795568 kbyte 2.After also having had a look to the manual I must admit that my memories of the DO model were rather simplified. I kept in mind only, that the DO model instead of ray racing considers a governing equation for radiation. Hence you are right, it is not only one equation but the number depends on the angle resolution for phi and theta. In my case I used the standard resolution 2X2 that means 32 additional equations for radiation had to be solved. Due to my observation that I needed 6.35% (this is the exact difference of the RAM requirements I gave in my last message) of the number cells as additional RAM in kbyte, you should need 2.25 times more (14.3%) due to your finer resolution. That means for your case of 4E5 cells a requirement of 57 MB. Does this confirm your observation? |
|
December 12, 2000, 08:07 |
Re: Memory needed for big case
|
#7 |
Guest
Posts: n/a
|
In fact, the typical memory need for a 2D model with turbulence standard k-eps model) is 1 Meg of Ram for 1000 cells. So for a test case: 1) 51 000 cells ~ 51 Meg of ram 2) 512 000 ~ 512 Meg If a DO radiation model and 2 additional species are added: 1) 88 Meg -> +72% 2) 719 Meg -> +40% We can assume that for a 800 000 cells problem, the additional Ram would be of the order of 30-35%
|
|
|
|
Similar Threads | ||||
Thread | Thread Starter | Forum | Replies | Last Post |
Memory needed for solving a 3 million model | eleazar | CFX | 5 | February 26, 2011 16:35 |
Changing the grid on the same set-up | Katya | FLUENT | 7 | October 8, 2009 17:31 |
icoFsiFoam case needed | vivekcfd | OpenFOAM | 1 | May 6, 2009 14:41 |
Transient case running with a super computer | microfin | FLUENT | 0 | March 31, 2009 12:20 |
Turbulent Flat Plate Validation Case | Jonas Larsson | Main CFD Forum | 0 | April 2, 2004 11:25 |