Wtf we all know that tons of code doesn't parallelize. That's why only a few applications benefit from vector extensions like AVX.

Even though RISC-V already managed to produce a few footguns ( like compressed instruction stream), I like it overall. It has became absolute smash arch for microcontrollers and it is rapidly rising elsewhere, from linux SoCs upwards.

But above all it is becoming the default base for all these exotic and esoteric new branches that Tenstorrent and so many others are going for.

I'd like to see someone doing smart vector/tensor unit that could do its execution autonomously, once its "shader program" gets filled. And I'd like to see that unit capable of doing its own conditional loops etc while the scalar core goes minding its own business.