hi,
I am *absolutly* not an expert in CFD...
I want simulate an air flow pressure in a pipe, with a disk simulating a valve movement.
I need to move the angle disk fonction of time.
So I would calculate the pressure evolution in front of the disk function of the time.
what version of code saturn I should use with a Unbutu 14.04 or 16.04 ?
I have a K40 nvidia processor.
How I do start to realise that ?
What are the difficulties ?
Laurent.
SIMPLE AIR FLOW VALVE SIMULATION (need some help)
Forum rules
Please read the forum usage recommendations before posting.
Please read the forum usage recommendations before posting.
-
LaurentDelaon
- Posts: 2
- Joined: Tue Apr 18, 2017 8:29 am
SIMPLE AIR FLOW VALVE SIMULATION (need some help)
- Attachments
-
- SchemaCFD_vanne.pdf
- Schema valve simuling.
- (85.97 KiB) Downloaded 270 times
-
Yvan Fournier
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: SIMPLE AIR FLOW VALVE SIMULATION (need some help)
Hello,
I recommend using Code_Saturne 4.0 or 5.0-beta (5.0 will be released by the end of this month).
You will need to recompile it. You can start using the Ubuntu packages instead (this will in any case make your future installation easier by pulling dependencies), but user subroutines do not seem to work using the Ubuntu package (they work in Debian), so you will be limited to the GUI.
For small movements, you could use a deformable mesh, but this will add complexity, and not work when the mesh is too deformed or topology changes, so I recommend building a series of meshes (one for each position, so scripting them is useful).
Code_Saturne does not directly handle GPU's (except optionally through PETSc, which assumes advanced user subroutines setup), so your NVIDIA GPU is less important than you host processor type.
Regards,
Yvan
I recommend using Code_Saturne 4.0 or 5.0-beta (5.0 will be released by the end of this month).
You will need to recompile it. You can start using the Ubuntu packages instead (this will in any case make your future installation easier by pulling dependencies), but user subroutines do not seem to work using the Ubuntu package (they work in Debian), so you will be limited to the GUI.
For small movements, you could use a deformable mesh, but this will add complexity, and not work when the mesh is too deformed or topology changes, so I recommend building a series of meshes (one for each position, so scripting them is useful).
Code_Saturne does not directly handle GPU's (except optionally through PETSc, which assumes advanced user subroutines setup), so your NVIDIA GPU is less important than you host processor type.
Regards,
Yvan
-
LaurentDelaon
- Posts: 2
- Joined: Tue Apr 18, 2017 8:29 am
Re: SIMPLE AIR FLOW VALVE SIMULATION (need some help)
Hi Yvan,
Thank's for your advertisings.
I have a computer like this :
Xeon E5-2620 v4 @2.1GHz
16 Go ddrm4 2133Mhz.
8 cores 16 threads
my question : is the cudaBlas not take into account ?
is code saturne using Blas or Atlas ? or a Library like that ?
thank's.
Laurent.
Thank's for your advertisings.
I have a computer like this :
Xeon E5-2620 v4 @2.1GHz
16 Go ddrm4 2133Mhz.
8 cores 16 threads
my question : is the cudaBlas not take into account ?
is code saturne using Blas or Atlas ? or a Library like that ?
thank's.
Laurent.
-
Yvan Fournier
- Posts: 4208
- Joined: Mon Feb 20, 2012 3:25 pm
Re: SIMPLE AIR FLOW VALVE SIMULATION (need some help)
Hello,
No, we dot not use BLAS libraries (except as an option for performance/precision comparison checks, and the Intel MKL as an option for sparse matrix-vector products), because Code_Saturne uses mostly BLAS-1 operations, for which BLAS libraries do not significantly improve performance (the arithmetic intensity being too low, these operations are bandwith-limited), but add complexity, especially when using OpenMP threads.
Best regards,
Yvan
No, we dot not use BLAS libraries (except as an option for performance/precision comparison checks, and the Intel MKL as an option for sparse matrix-vector products), because Code_Saturne uses mostly BLAS-1 operations, for which BLAS libraries do not significantly improve performance (the arithmetic intensity being too low, these operations are bandwith-limited), but add complexity, especially when using OpenMP threads.
Best regards,
Yvan