can these do training or only inference? currently working on learning machine learning and I'd love to have a physical machine I could aim to build real workloads on in a few years.

▲

LogicFailsMe 2 hours ago | parent | next [-]

They're Turing complete. What else do you need?

▲

Npovview an hour ago | parent | next [-]

There is a reason why Google has tpu8i and tpu8t

▲

porphyra 2 hours ago | parent | prev | next [-]

technically in order for something to be turing complete it needs infinite memory

	▲	LogicFailsMe an hour ago \| parent [-]
		The more (memory) you buy, the more you save!

▲

Our_Benefactors 2 hours ago | parent | prev [-]

It’s possible (likely, even) to have a chip fast enough for inference, but not fast enough or with enough memory to do meaningful training runs. Like the current DGX spark.

▲

airjason 2 hours ago | parent | prev | next [-]

not for llm full training, but can do some finetuning for sure.

▲

the_real_cher 2 hours ago | parent | prev [-]

I believe training is way more processor intensive than inference.