these are much cheaper to produce and run, not to mention more flexible than GPU’s for LLM training. as an added benefit you can write (if you were so inclined)

these are much cheaper to produce and run, not to mention more flexible than GPU’s for LLM training. as an added benefit you can write (if you were so inclined) hardware-level logic to whatever workload you want. reprogrammable as well. -background : 4-5 years ago worked on embedded systems tech that used these for fintech. they seemed like the future back then and LLM workloads were barely a thing. Glad more people are learning about them.