IRTools.jl VS Dagger.jl

Which is the concept behind Cassette.jl (https://github.com/jrevels/Cassette.jl) and IRTools.jl (https://github.com/MikeInnes/IRTools.jl).

Firstly, I'll say that we already have work started to implement out-of-core directly in Dagger: https://github.com/JuliaParallel/Dagger.jl/pull/289.

With that PR in place, it should be possible to define a "storage device" which is backed by a database. I haven't had a chance to actually try this, since the PR still needs quite some work and testing, but it's definitely something on my radar!

I'm not sure that's necessarily the domain of a low-level package like CUDA.jl though (which I assume you're referring to). That kind of interface is more the domain of higher-level packages like https://github.com/JuliaParallel/Dagger.jl/ and to a lesser extent https://juliagpu.github.io/KernelAbstractions.jl/stable/. Moreover, the jury is still out on whether the built-in Distributed module is an ideal abstraction for every use-case (clusters, heterogeneous compute, etc.)

WRT Nx, my biggest question is how they'll crack the problem of still needing big balls of C++ and the shims everywhere to get acceleration. Creating a compiler that generates efficient GPU or other accelerator code is a massive research project with no clear winners, never mind the challenge of reconciling the very mutation-heavy needs of GPU compute with a mostly immutable language model.