3 things to know about Ironwood, our latest TPU
15 hours ago
- #Google Cloud
- #AI Hardware
- #TPU
- Ironwood is Google's seventh-generation Tensor Processing Unit (TPU), designed for speed and efficiency in AI models.
- It offers more than 4X better performance per chip for both training and inference workloads compared to the previous generation.
- Ironwood can scale up to 9,216 chips in a superpod, linked via a 9.6 Tb/s Inter-Chip Interconnect (ICI) network.
- The TPU features 1.77 Petabytes of shared High Bandwidth Memory (HBM), reducing compute-hours and energy for AI services.
- Ironwood's design is influenced by Google's AI research, including the use of AI (AlphaChip) to optimize chip layouts.