1 of 1
Story summary
- Google launched eighth-generation Tensor Processing Units, splitting AI training and inference into TPU 8t and TPU 8i.
- The change aims to boost performance and cost efficiency amid rising AI demand.
- The TPU 8t delivers nearly three times the performance of its predecessor.
- The TPU 8i provides 80% better performance per dollar for inference.
- Google and NVIDIA remain in competition, with the new chips launching later this year.
