No .f90 in hemi and turbChannel examples

[timofeymukha]

Basically, subj.
For hemi there is one here
https://github.com/ExtremeFLOW/test-suite/blob/master/neko_hemi
can it be reused?

[njansson]

The case in src/examples/hemi is an example of "standalone" case whereas inflow etc is all defined in the core solver.

To run it simply copy the neko binary from the src (or from bin under the installation prefix) and run with neko hemi.case

However, the current version of this case is slightly different from the test suite. As default it uses a uniform inflow profile, but it could be changed by adding/modifying the following in the case file,

&NEKO_CASE
...
initial_condition = 'blasius'
/

and

&NEKO_PARAMETERS
...
fluid_inflow = 'blasius'
/

there's also a parameter delta and blasius_approx under NEKO_PARAMETERS related to the boundary layer thickness (default 1) and the profile approximation (default sin).

[timofeymukha]

Ah, I see. Nice to be able to run "standalone". I guess with modern fortran we can go quite far in terms of allowing to specify everything in the config instead of tinkering in user-defined functions.

[njansson]

Correct, that is at least the intent. Of course one can't cover all the possibilities, but it will make life much easier for users on accelerators if "standalone" would cover most of their cases (otherwise people would need to write accelerator code in their user file)

[timofeymukha]

@njansson why does the channel case not use avflow, vol_flow_dir, flow_rate to drive the flow? On the other hand, uinf is set, what does that do exactly?

[njansson]

Definitely due to neglect from our side, the case is not setup correct afaik. @MartinKarp can correct me but I don't think we had vol_flow_dir implemented when we use that case to debug periodic conditions.

Ps. vol_flow and friends are currently only available in the plan4 solver. The pipe case is a good example on how to use it

[MartinKarp]

Yeah, a forcing or volflow is necessary for the channel case.

[njansson]

Patches are more than welcomed (^-^)

[timofeymukha]

[njansson]

vol_flow should now work with the fluid_pnpn solver. A device version will appear soon, progress can be tracked in #554

[Bkv-1915]

HI. I am trying to run the turb_channel example on my GPU. However, it says that it will run in 10 minutes. I have RTX 4090 in my workstation and it takes 150 minutes. Am I missing something?

[timfelle]

Which command did you use to setup Neko and which command did you use to execute the example?
This sounds to me like you are either not using the gpu or you are launching more than one mpi ranks which will then compete for using your gpu.

[Bkv-1915]

I used : ./configure --with-cuda=/usr/local/cuda --prefix=/path to setup Neko. To execute the example, I used: mpirun -np 24 ./neko turb_channel.case. I did this after /path/bin/makeneko turb_channel.f90. FYI I am new to this. I am sorry if there is a basic mistake here.

edit: I checked the processes and it says 100% GPU and 99% CPU is being used while the simulation is runnig.

edit2: Its way quicker when I do mpirun -np 1 ./neko turb_channel.case. Thanks @timfelle

[timfelle]

Great, I am happy that helped resolve the issue. I will also just take a look at our documentation and see if we actually are clear enough about that little detail 😅 It might be one of those that have slipped through the cracks.