Hahahahahaha, 'MVP' doesn't exist in hardware or low-level hardware coding (e.g. this, which is a playstation game that presumably runs bare-metal).
In hardware, it's generally like ~80% of the design work before you even have a viable demo. If you're writing a bare-metal rendering engine, it's very much the same. MVP isn't used because there is none of the infrastructure even available to have one.
TL;DR MVP isn't possible if you don't have the infrastructure.
Not sure I agree. It is certainly possible to hack up the basic functionality of an embedded system; full of magic numbers and lacking the elegant hardware abstractions, OS layer and power management we all want.
Edit: Not to mention in hardware, where you can neglect design for manufacturing, solder modwires everywhere, and power everything off a 10kg power supply. MVPs do exist in hardware and embedded SW.
I think there's a difference between an MVP and a proof of concept.
You can hack together a one-off PoC for a hardware feature, in order to know the feature is viable--but what you've created isn't a viable product, minimal or otherwise. With an MVP, you'd expect to be able to incrementally improve the design into a non-minimal one, but with a PoC there's no real way to take it and "refactor" it into a shipping product at all. The design with modwires and a 10kg power supply (or the embedded-software equivalent) doesn't share any similarities, except at the most abstract level, with the design that can be mass-manufactured.
Once you're done experimenting and you know what you want, you really do need to design the "production model" from the top down, with all the layers that will enter into the solution considered in advance.
A concrete example: you can't just "write a voxel renderer" and then expand it into a full game engine. If the full game engine will have a rendering pipeline with support for things like dynamic lighting and fur physics, then voxel rendering will be implemented in that engine in a completely different way than in one that only needs to render voxels.
Editing one into the other would be like editing the code defining a b-tree into code for a hash-table: there would be exactly one edit, which would replace all the old lines with entirely new ones.
In hardware, it's generally like ~80% of the design work before you even have a viable demo. If you're writing a bare-metal rendering engine, it's very much the same. MVP isn't used because there is none of the infrastructure even available to have one.
TL;DR MVP isn't possible if you don't have the infrastructure.