Learning Forum

[Sam001]

Hi,

   I have a mesh (FFF.msh) size of 65GB, I tried lauching fluent with it, hardware RAM of 768G into 16 cores.

I discovered I had errors shown in the pic1 below, when I run top, I discovered the fluent software runs out of memory for GUI.

I tried the same thing via command line using (16X64GB RAM) but had errors in pic2.

Please, is there a workaround this?

Also, is there a X value of mesh for Y value of Fluent?

Furthermore, does anyone know how to write codes for mesh improvement in order to achieve a determined orthogonal quality in fluent?

 

Thanks,

 

Sam 

 

 

[Keyur Kanade]

what is cell count?

[Sam001]

approximately 895 million 

 

 

[Amine Ben Hadj Ali]

Try using larger memory machine. Rule of thumb 1 KiloByte for one cell..

[Sam001]

Amine,

          For 895 million cells multiply by 1 kilobyte = 895 GB, on the second pic of each post, I used 992GB(allowable in 16X64GB) machine which is like 80GB+ more(I explained that in my first post).

Thanks,

Sam

[Rob]

Looking at the error Fluent is  looking for about 1700 GB?  The older rule of thumb was 1M cells per GB RAM but with the newer numerics it's nearer 2GB now. 

[Sam001]

rwoolhou, but I thought Ansys said they would be reducing memory (RAM) required in this newer Releases? 

[Amine Ben Hadj Ali]

To add to my colleague's post: it is case and models dependent. What are you modelling?

[Sam001]

Amine, compact heat exchanger

[Amine Ben Hadj Ali]

So I guess then energy equation, at least two cell zones, double precision you are then very quick out of memory for this number of cells at this moderate number of CPU's and moderate RAM. Either you reduce the count or you extend your hardware.

[Keyur Kanade]

you really need to reduce the cell count. 

do you really need so much cell count for a compact heat exchanger? 

[Sam001]

kkande, my geometry is very small (12"X3"X0.6") with tiny standing fins stacked inside, Am currently using non-conformal meshing, if i reduce my cell count(if i understand you well, you mean reduce the mesh size), i won't be able to achieve an mesh orthogonal quality of 0.79+ (implying divergence).

 

[Rob]

Please can you post some images of the HX geometry.  If you're including fins etc on one side of the system you may need to reduce the amount of domain modelled (potentially using periodic conditions). I'm at about 30M poly cells to model three passes (two hot side & one code side) of a plate HX: it should be about 15 passes each way. 

[Sam001]

rwoolhou, am not even on passes yet but a single flow through the fin with heat flux application on top and bottom plates, also with insulation on side plates. Its worth of note, the fins are inside the box (standing stacked along).

 

[Amine Ben Hadj Ali]

Model a block if fins and make use if symmetry BCs ans/or translational periodic ( if the latter are appropriate). Other alternative us tivuse one of our heat ecgsnger models where the HX is not resolved.