It's so funny to see FP4.... I remember 20 years ago being asked what sort of HPC we needed in genomics, and the answer was basically, "lower precision, faster" for the stuff I was working on. But FP4 is, well, almost comical.

One thing not on that comparison table: die size. If I'm understanding that correctly, it's about the same as the Rubin, but at 1/3 the number of NVFP4 PFLOPs. (The text disagrees with the table, I'm taking the table as truth, perhaps that's wrong...)

▲

nxtfari 7 hours ago | parent [-]

Agree, I remember when even half precision made its way into C# sometime around 2020 (I didn’t know much about ML then) and I thought, well I guess that’s a worthwhile tradeoff but I can’t imagine going lower. Lo and behold (1-bit Bonsai) how much lower you could go.

▲

jacquesm 6 hours ago | parent [-]

Ternary?

▲

jeffbee 5 hours ago | parent [-]

Knuth's base-e proposal enters the chat.

They were right about everything 50+ years ago, but they didn't have the budget for the right hardware, had to write conference papers and books instead.

▲

jacquesm 4 hours ago | parent [-]

I can totally see how ternary would work from a physical implementation perspective but I have a really hard time visualizing anything using base-e, can you explain how such a thing would work in practice?

▲

jeffbee 4 hours ago | parent [-]

No it's impossible. But it would be optimal!

	▲	jacquesm 3 hours ago \| parent [-]
		Ah, the spherical cow of number bases :) Thanks for the response, that saved me a sleepless night.