I don't think it does? > Google's v5p TPUs are up to 2.8 times faster at training large language models than TPU v4, and offer 2.1-times value-for-money. That

I don't think it does? > Google's v5p TPUs are up to 2.8 times faster at training large language models than TPU v4, and offer 2.1-times value-for-money. That's a comparison of TPUv5p to TPUv4, not to H100.