TPUs
Web page · https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx
First seen by Oriel on .
AI summary
Google introduces TPU 8tx and TPU 8i Max, expanding its eighth-generation portfolio with systems for confidential inference and massive-context workloads, alongside an open software stack for agentic AI.
- TPU 8tx targets regulated enterprises and governments, protecting active data and model weights through hardware-level confidential computing.
- Using TPU 8t silicon, 5th Generation AMD EPYC CPU hosts, and Boardfly networking, TPU 8tx delivers up to 2x lower latency than TPU 8t with near-zero performance penalty.
- TPU 8i Max combines two low-latency TPU 8i chips, doubling compute, SRAM, and HBM bandwidth while fitting twice the silicon into the existing rack footprint.
- TorchTPU runs PyTorch code and Hugging Face models natively on TPUs, while vLLM, SGLang, TPU agent tools, and XLA Serving Engine broaden deployment options.
- TPU Sync and GKE dynamic slicing address large-context memory and fleet utilization, cutting Time to First Token by over 90%, startup times by up to 5x, and fault recovery by up to 4.5x.
#tpu #agentic ai #confidential computing #pytorch
Generated by gpt-5.6-sol. Check the original for details.
Metric trends
Last 30 days · Up to 120 recorded samples per source and metric, with reposts kept separate · First and latest retained
Points
Score reported by the source.
- Hacker News: 254, latest observation ; +228 over observed period since
Comments
Comment count reported by the source.
- Hacker News: 75, latest observation ; +74 over observed period since
Aliases
| Source | External ID | URL |
|---|---|---|
| Hacker News | 50432831 | https://blog.google/innovation-and-ai/infrastructure-and-cloud/global-network/tpu-8i-max-and-8tx/ Read discussion |