I have a basic CFX simulation already set up. I have lots of parameters I need to vary making the simulation take much longer than we have time. I want to speed up the simulation, but have no idea, could anyone make a suggestion please ?
Hi Alireza, you can run this calculation with a high config systems, If it is impossible, you can reduce mesh number with use of larger meshes in geometry.
good luck
Dear alireza Ansys CFX uses the control volume finite element method, No Finite volume method, you can tag this expression to your question.
good lock
Hi Ali,
You can allocate system memory resources in CFX Solver to increase your solver performance and decrease computation time. By default it is set to 1 and you can observe that not all of your system memory/RAM is engaged in CFX calculation. You need to be very careful in increasing memory allocation factor in CFX Solver window. I mean you have to check and review; Let say you put 1.1 and then you have to check your system memory performance as it is allocating your system memory resources. Then you can change it to 1.2 and then review. Don't put a high value until you are 100% sure about its effect on computation performance.
-Faizan
How complex is your problem? Using a better/physically meaningful initial condition could be a good idea for a highly non-linear problem.
Hi Ali,
generally, improve your mesh quality and reduce the number, would be helpful to reduce computing time, especially for complex object.
Hello Sayed,
There are some points where you can work:
- Time step: rarely the automatic time step is a good choice to complex problems. It is calculated based on the dt = L/U and L is L= (volume of domain )1/3 . It is a good guess, but generally you must try other values - you will be face with precision vs. computational time. If it is a transient run, a good time step is that remains the internal time looping 1
Hi,
essentially the best speedup can be achieved by:
1) brute force, i.e. using massive parallel computational resources for the simulations. On a large enough parallel cluster system even a manifold of parameter analysis can be accomplished in very short time.
Since answer1) might be not that helpful because of licensing implications and because cluster resources might be limited or not available at all, so here my suggestion 2):
2) Get to the epicenter of your parameter space and carry out for one selected parameter set the corresponding CFX simulation with care.
Once you are confident about this result, take the RES file as an initialization for all further parametric and mesh and other studies. If the "master case" has been selected carefully, than the predicted flow will have quite some similarities with the flow solutions for the other parametric studies, hence the flow variations from this already quite good initial guess towards the desired solutions is much less than from any other initial guess which you might use otherwise. Consequently this potentially allows you to arrive at the CFX solutions in substantially less time (e.g. by being allowed to use larger timescales, less underrelaxation, smaller number of iterations,...).
Regards,
Dr. Th. Frank.
Naseem Ahmad
: In my honest opinion, the amount of provided memory is not changing the number of numerical operations which need to be solved / calculated to find the solution of a numerical CFD problem. It stays the same with no regard to the memory amounnt and therefore on 2 computers with differnt amount of memory the solution time should be equal. From this perspective I cannot really see, how an increase in memory provision can lead to a quicker CFD solution in ANSYS CFX or in any other CFD code. The only exception is, if your memory is so small, that your computer starts to use e.g. on Linux the swap partition of your hard disk. But than the computation becomes extremely slow and essentially than either the mesh is too large or the computer not capable to be used for CFD.Other answers in this chat are providing better hints on how simulations can be speed up.
Regards,
Dr. Th. Frank.
Naseem Ahmad
: Again. Adding RAM is not speeding up any computation. If the number of computational instructions stays the same, it will be equally fast computed on a system with less or with more RAM. Only if the software is parallelized and the number of computational instructions is carried out by multiple processor cores, than the work can be accomplished in less time. But this has not so much to do with system memory, i.e. RAM.
Naseem Ahmad
: What do you mean by "specifications"?Isn't it obvious, that the CPU is doing the calculation work and not the memory of a system?
Agreed with Thomas. "RAM is not speeding up any computation or can not reduce the computation time."
I reckon time step function can reduce the simulation overall computation time. Also check with residuals and try to achieve the convergence.
In my opinion, processor is more important than RAM. You may have 8 or 16 GB RAM but if your processor contains lesser cores you might not be able to run simulation faster.
I am agree with Thomas Frank and Muhammad Faizan-Ur-Rab
Thank you,
- Badges
- Science topic