@bob I've been mulling this over a bit. Like, "linear algebra machine" seems like a pretty obvious one, I guess with a bias towards DSP wot makes convolutions go brrrr
but then, that's what GPUs are.
I'm realizing as I write this that I don't know enough about the silicon/compute architecture […]
@SnoopJ they vary in terms of memory architecture, the only commonality is they have big systolic matrix units. some are VLIW machines that look like DSPs (qualcomm, groq), some are cgra-style grids (sambanova, tenstorrent), some are GPUs. the main difference between all of these and GPUs from […]
Comments
but then, that's what GPUs are.
I'm realizing as I write this that I don't know enough about the silicon/compute architecture […]
wait a minute
the point about scheduling is… one I'm going to have to think about some more.
…graph theory machines?