What cells should I use in CFD? | Tetrahedral vs Mixed cells

Nice tests. Thank you for sharing.

[@Thierry080]

Interesting, thanks for sharing. How transferrable are the results to other solvers? Would the conclusion hold up in, for example, an FEM solver instead of FVM? What about higher order cells? The potential speed up achieved with a mixed mesh vs. pure tets is of course always welcome, but do you have an educated guess about the result performance?

[@interfluo6420]

Like I said in the video it is pretty difficult to make sweeping statements, it is possible to try to make educated guesses but the effects from particular implementations cannot be ignored. Generically I can say that tets are typically not used for 1st order FEM and are used a lot in FVM & higher order; this is more of an observational statement and not really a guess though. One thing which cannot be neglected is engineering time, it is super quick to throw a bunch of tets at a problem if you have the computational resources, this is often a decent practical approach.

[@AwestrikeFearofGods]

So...tetrahedral mesh had 2.36x as many cells, 0.99x as many nodes, and 1.62x as many edges...as mixed mesh. That's a big deal if run duration was only 0.73x! Why isn't this common knowledge already? Good work! Did you run exactly 10,000 iterations for each case? It appears so from the charts, but it's hard to tell with a logarithmic domain and holes in the data. 2:11 On the right-hand chart, what the hell's going on with the y-axis? Did you round the values to the nearest tenth?

[@nicolasflamel6518]

thanks for sharing. Is the geometry file public?

[@interfluo6420]

The geometry file is not public unfortunately

[@oskarelmgren]

A lot of conservative CFD people are going to be very mad at these results. Not because the results are wrong, but because they won't like the results. :D