Watch Desk posted an update
Alphabet’s latest TPUs offer up to 80% better inference price-performance than their predecessors, according to Seeking Alpha contributor Brett Ashcroft Green. He also says they deliver up to 2.7 times better training performance, with nearly 100 customers using the chips.
Why it mattersThat makes this a useful signal about the hardware choices behind AI services: specialised chips could help serve models more efficiently, while Alphabet is expanding its customer base. The figures are claims in an investment analysis, not an independent benchmark. Ashcroft Green disclosed that he holds shares in Alphabet, Nvidia and Amazon.
Discuss: If specialised chips can make AI inference cheaper, should providers invest in them even if that means relying less on versatile GPUs?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.