

Well yeah sure if you want a set algorithm to perfectly reproduce this exact universe deterministically, that’s not gonna work out so well.
But a simulation doesn’t have to be perfectly consistent and deterministic to “work”. If anything, the fact that some things can’t be predicted is evidence in favor of us being in a simulation, not against.
This paper just rules out a class of algorithms. Were not in a specific type of simulation. Doesn’t mean we’re not in a simulation at all.

I have been doing a bit of compute work on nixos with both AMD and nvidia, and I’d say it depends on what you’re doing.
If you’re doing your compute via compute shaders, you’ll have a great experience on AMD. Zero hiccups for me, I just wrote my shaders and ran them no problem. Vulkan is incredible.
If you have to interact with other people’s compute crap though, it might be a bad time. Most folks do GPU compute with cuda, and that won’t be fun for you on AMD. Yes there are translation layers, and you can make them work for some use cases, but its a bad experience. And yeah rocm exists… but does it really? Not many cards actually support rocm, and software support for it is just as sparse.