Google expands eighth-generation TPU lineup with two systems

Google has announced two new eighth-generation TPU systems, TPU 8tx and TPU 8i Max, expanding its TPU portfolio to four systems in a single year.

• TPU 8tx is designed for secure and confidential inference and small-scale training, using the same silicon as TPU 8t and 5th Generation AMD EPYC CPU hosts; it aims to protect sensitive data and model weights in active use.
• TPU 8i Max combines two TPU 8i chips in one package for massive context windows, delivering double compute, double SRAM, and double HBM bandwidth versus TPU 8i, with higher compute density per rack.
• Google also highlighted recent software and ecosystem updates including TorchTPU, XLA Serving Engine, Agent Compute Engine, TPU Sync, Cloud Pathways, and GKE dynamic slicing; the post references a 9 a.m. PT broadcast for more details.