The speed of WUFI 2D simulations

All about WUFI 2D
Post Reply
Joseph Little
WUFI International Support Team
WUFI International Support Team
Posts: 13
Joined: Wed Dec 15, 2010 6:25 am -1100
Location: Dublin, Ireland
Contact:

The speed of WUFI 2D simulations

Post by Joseph Little » Wed Sep 07, 2011 6:19 am -1100

Hi,
I have been told I have the right processing power and RAM on my computer for WUFI 2D (given the stated system requirements), but I’m finding it SO slow! One simulation is now running four hours and the computer tells me there is another 4 hours to go! I made commitment to a client I can’t now meet because of this.

1) Do you know if it’s possible to take less than 8760 steps in simulation?
2) Does WUFI 2D know each step is an hour? Could we have two hour steps for instance?
3) Does it matter if ‘outside’ is top of screen rather than left hand side? (In the current very slow simulation the former is the case).
4) Is it always essential to run initial conditions in WUFI Pro first to feed in considering we're only running one year (even in 'dry' buildups like roof)?
5) Any advice to speed it up?

Many thanks.
Joseph
:(

Christian Bludau
WUFI SupportTeam IBP
WUFI SupportTeam IBP
Posts: 1130
Joined: Tue Jul 04, 2006 10:08 pm -1100
Location: IBP Holzkirchen, the home of WUFI
Contact:

Post by Christian Bludau » Wed Sep 07, 2011 7:31 pm -1100

Hello Joseph,
first I have to say, yes, Wufi 2D needs its time. But if you use a not to huge construction and an appropriate grid, it can speed up things. Unfortunately thats not the only influence to the speed. Also the composite of the construction, the initial conditions... have an influence. But to your questions:

1)2) It is possible to use bigger time steps. For that go to the "Enhanced"menu in Computational Parameters. Here you can change the time step. 2 hous would be 7200s. But keep in mind that this will have an effect on the accuracy of the results, mainly the relative humidity.

3) No, there is no influence of the orientation of the boundary conditions. It more could be the boundary condition itself. For example if you have a high number of days with precipitation, it slows down the calculation, because you need more iteration steps (rain iteration) for every time step.

4) I do that really seldom, only in case of special question.

5) Yes. Try to brake down your construction to a as simple and as small as possible model. Cut the constructions at symmetric axes. Do not simulate a 10meter roof section, only because it is 10meters long. Cut in the middle of the mineral wool on one side and in the middle of the rafter on the other side for example.
Further the grid is important. Do not make it too fine, it will take much time but also do not generate it too small, the results will be not good. A steady-going grid will show better convergence behavior, than a size-jumping-grid. The expansion factor should not be higher than two respectively lower than 0.5.
There is also a influence of the used materials in the construction. If there is a heavy bend in the moisture storage, convergence may be much harder than with a steady-increasing function (maybe generate it).

But for short: The best way for a faster simulation is to keep the construction as simple as possible and the grid as small as acceptable. I would not screw around with the time. Better set up your construction during day and let it calculate during night.

If you use an notebook, keep in mind that some settings may slow down the processor during you are not working (only calculating) with it to save energy.

Hope that helps,
Christian

olegh
WUFI User
WUFI User
Posts: 2
Joined: Thu Feb 28, 2019 10:31 pm -1100

Re: The speed of WUFI 2D simulations

Post by olegh » Mon Mar 25, 2019 11:46 pm -1100

Hi!
I have also encountered some difficulties with simulation speed in WUFI 2D. Especially with WUFI 2D 4.1.
Using WUFI 2D 4.1. a simulation took 116 hours. While using WUFI 2D 3.4. for the same model, the simulation only took a couple of hours.
The results look to be the same when using version 4.1 and 3.4.
Does 4.1 implement anything special that will have an effect on the results that wufi 3.4. does not? Or is the difference between the versions mainly the GUI?

Thanks in advance,
Ole G

Christian Bludau
WUFI SupportTeam IBP
WUFI SupportTeam IBP
Posts: 1130
Joined: Tue Jul 04, 2006 10:08 pm -1100
Location: IBP Holzkirchen, the home of WUFI
Contact:

Re: The speed of WUFI 2D simulations

Post by Christian Bludau » Tue Mar 26, 2019 1:14 am -1100

Dear Ole,
WUFI 2D 4 is able to use to use all calculation cores, what WUFI 2D 3 is not doing. So in most cases it should be faster than WUFI 2D 4.
The following setting would have an high influence on the speed:
- convergence criteria
- number of grid elements
- adaptive time step control
- and others.
Are you sure this settings are all unchanged?
Christian

zhanqsh
WUFI User
WUFI User
Posts: 11
Joined: Wed Nov 13, 2019 4:23 am -1100

Re: The speed of WUFI 2D simulations

Post by zhanqsh » Thu Apr 02, 2020 9:57 pm -1100

Hi,

I met a wired problem with the simulation speed in WUFI 2D-4.

I ran the exact same model with different processing powers and RAM on a supercomputer for WUFI 2D (given the stated system requirements). But when with 24 cores CPU and 48G RAM, the estimated running time was more than 10 Days; when with 8 cores CPU and 16G RAM, it changed to 6 days; when with 2 cores CPU and 4G RAM, it changed to 16 hours.

Isn't WUFI 2D-4 able to use all calculation cores? Is there any necessary settings when using multi-cores for simulation?

Thank you!

Sebastian Boegl
WUFI SupportTeam IBP
WUFI SupportTeam IBP
Posts: 46
Joined: Sun Jan 06, 2013 11:45 pm -1100
Contact:

Re: The speed of WUFI 2D simulations

Post by Sebastian Boegl » Sun Apr 05, 2020 10:55 pm -1100

Dear zhanqsh,

By default, WUFI 2D-4 uses all but one CPU core to calculate the simulation.
As more cores require a higher synchronization effort, more cores are not always faster. Depending on the selected grid, 2 to 4 cores show fast results.
You can easily change the desired number of parallelization by setting the process environment variable

Code: Select all

OMP_NUM_THREADS=<NUMBER_OF_CORES + 1>
e.g. to use four parallel computations start WUFI 2D-4 with:

Code: Select all

SET OMP_NUM_THREADS=5
If you run your simulation in batch mode you should directly set the number of cores, i.e. not incrementing it by the one core that uses WUFI2D for the GUI.
Avoid setting the value larger than the number of available cores.

Sebastian

zhanqsh
WUFI User
WUFI User
Posts: 11
Joined: Wed Nov 13, 2019 4:23 am -1100

Re: The speed of WUFI 2D simulations

Post by zhanqsh » Wed Apr 08, 2020 1:57 am -1100

Dear Sebastian,
Thanks for your reply.
BTW, where can I set the process environment variable?

Thank you very much!

zhanqsh

Sebastian Boegl
WUFI SupportTeam IBP
WUFI SupportTeam IBP
Posts: 46
Joined: Sun Jan 06, 2013 11:45 pm -1100
Contact:

Re: The speed of WUFI 2D simulations

Post by Sebastian Boegl » Wed Apr 08, 2020 2:47 am -1100

Dear zhanqsh,

Either create a shortcut to open WUFI2D and change the destination to

Code: Select all

C:\Windows\System32\cmd.exe /c "SET OMP_NUM_THREADS=4 && START /D ^"C:\Program Files (x86)\WUFI\WUFI2D-4\^" WUFI2D-4.exe"
or simply by invoking the command within a cmd (windows command prompt).

Regards,
Sebastian

zhanqsh
WUFI User
WUFI User
Posts: 11
Joined: Wed Nov 13, 2019 4:23 am -1100

Re: The speed of WUFI 2D simulations

Post by zhanqsh » Wed Apr 08, 2020 6:09 am -1100

Dear Sebastian,

Thank you! It works.

Post Reply