"Ironwood delivers 4.6 petaFLOPS of peak FP8 compute per chip, roughly four times the performance of its predecessor Trillium, with 192 gigabytes of HBM3e memory and 7.37 terabytes per second of memory bandwidth."
"A single Ironwood superpod links 9,216 chips into a unified system delivering 42.5 exaFLOPS of compute, more than 24 times the capacity of El Capitan, currently the world's most powerful supercomputer."
"Google's advantage is at cluster scale: the superpod architecture, its energy efficiency at roughly twice the performance per watt of Trillium and 2.8 times that of Nvidia's H100."
Google launched its seventh-generation TPU, Ironwood, at Cloud Next 2026, offering 4.6 petaFLOPS per chip and 42.5 exaFLOPS in a superpod. The TPU 8t and TPU 8i, designed for training and inference respectively, are set for late 2027. Ironwood's performance is four times that of its predecessor, Trillium, and it features 192 gigabytes of HBM3e memory. The architecture positions Ironwood as a competitor to Nvidia's Blackwell B200, with advantages in energy efficiency and cluster scale despite Nvidia's superior single-device interconnect bandwidth.
Read at TNW | Deep-Tech
Unable to calculate read time
Collection
[
|
...
]