Post #4146372
2026-07-27 19:32 UTC
@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
Replies (1)
-
@datenwolf@chaos.social 2026-07-27 20:33
@anthropy@mastodon.derg.nz @Em0nM4stodon@infosec.exchange Only recent versions of the tensor cores do support the low precision formats. OG tensor cores (and TPUs) for that matter, would have best performance with fp16 but also support "pseudo" fp32 (same range but less precision) and even fp64 albeit with much reduced performance. These are the useful number formats for scientific and simulation stuff like weather and climate modeling. The low precision formats are a rather new addition and pretty useless for DSP & sim.