Google announced the next generation of TPUs at I/O conference, Trillium, boasting 4.7x performance boost compared to the fifth generation, with enhanced MXUs and memory bandwidth.
Trillium introduces SparseCore accelerator for faster model training and lower latency, aiming at advanced ranking and recommendation workloads, claiming to be Google's most energy-efficient TPUs yet.
Pichai emphasized the importance of energy efficiency in meeting the surging demand for ML compute, stating that the new TPUs are 67% more energy-efficient than the previous generation.
[
add
]
[
|
|
...
]