AMD and Cerebras Launch AI Inference Solution

techgnosis 2 days ago | parent | next [-]

I'm not sure I understand. It's Helios but with WSE attached? And it uses one or the other depending on some criteria?

	▲	wtallis 2 days ago \| parent [-]
		LLM inference is a two-phase process. The first phase is prompt processing aka prefill. It's compute-heavy but requires relatively low memory bandwidth. The second phase is token generation aka decode, which doesn't require much in the way of FLOPs but wants as much memory bandwidth as possible. This announcement is for a system to do the first phase on Helios and the second phase on WSE.

JSR_FDED 2 days ago | parent | prev | next [-]

It’s disaggregated so you Know it’s good

	▲	rbanffy a day ago \| parent [-]
		The two major sides of the system excel at different things that happen to be complementary in AI inference.

ckrapu 2 days ago | parent | prev | next [-]

They're a bit late to the party.

	▲	rbanffy a day ago \| parent [-]
		Why would you say that?

sccvcxv 2 days ago | parent | prev [-]

lol when is the music gonna stop?

	▲	rbanffy a day ago \| parent \| next [-]
		Nobody knows the answer to that, but if it’s efficiency they are aiming for, which results in lower inference costs, it’ll last for a while. The inference to training ratio will only get higher and the margins will only get lower.
	▲	throw1234567891 14 hours ago \| parent \| prev [-]
		When Etched becomes GA.