Tomás Chor

> @tomchor Not sure if this is the issue but have you tried adapting `ScaleInvariantSmagorinsky` for the GPU? > > Looks like `ScalarBiharmonicDiffusivity` needed it here: > > https://github.com/CliMA/Oceananigans.jl/blob/b3be950ef6a1f2139c2f65e083f20f1f7958ea55/src/TurbulenceClosures/turbulence_closure_implementations/scalar_biharmonic_diffusivity.jl#L109-L113 >...

I wrote some validation scripts and I think the model is working, but I'd appreciate some feedback. First, this animation shows two 3D decaying turbulence simulations, one with `SmagorinskyLilly` and...

With the latest changes the compilation time is much shorter (around 10x shorter on the GPU) and simulations have been running in about twice to four times as long as...

I'm surprised at how much `SmagLilly` kills turbulence at these resolutions. To the point that $w^2$ is pretty much zero. A visualization of that effect for a wall flow can...

> I can't see the image because it seems to be protected. Weird. It's the second video in [this comment](https://github.com/CliMA/Oceananigans.jl/pull/3642#issuecomment-2417879667) in this PR.

Here are results including the Lagrangian-averaged version of the dynamic Smagorinsky with $\Delta z=2$ m:  As expected, for this simulation the results are pretty similar to the directionally-averaged version.

> Lagrangian-averaged should be slightly less diffusive right? Hmm, I'm not aware of that. afaik it's about the same in a doubly periodic domain. Looking at Bou-Zeid's Figure 1 I...

> I disagree that fig 1 does not have an obvious trend, the trend is quite clear away from the wall: > > Both Lagrangian schemes have larger coefficients than...

~I've been getting the following error right at the end of docs on buildkite that I can't reproduce locally (and therefore have not been able to solve):~ ```julia ┌ Error:...

FYI the tests are failing only because somehow tests are getting a kill signal or the servers are being lost:  Maybe something's up with he servers? As far as...