Post #4146373
2026-07-27 19:25 UTC
@anthropy@mastodon.derg.nz @Em0nM4stodon@infosec.exchange I'm not talking about TPUs. I've been programming them myself. I'm talking about the hardware being optimized for fp8, fp4 or just fp2 in some cases. Total garbage except for modeling neutral networks (sans the activation function).
And I'm talking about the bonkers rate of depreciation.
Locally executed machine learning models? Fine by me if the training isn't setting the planet ablaze & attempts at compensating the diminishing returns by throwing more resources at it.
Replies (1)
-
@anthropy@mastodon.derg.nz 2026-07-27 19:32
@datenwolf@chaos.social @Em0nM4stodon@infosec.exchange TPUs are "hardware optimized for FP8/FP4/FP2/INT8" and matrix calculations. Tensors, Scalars and Vectors, hence the name Tensor Processing Unit. Nvidia just calls them "GPUs" because they don't want to be confused with Google's TPUs. But technically, TPU is the specialized hardware you're referring to. And this underpins much of today's technologies, so I'm not really that upset about the hardware, or even LLMs. I just hate the needless unsustainability and plagiarism