Search
Google’s Eighth-Gen TPU Introduces Dedicated ‘Training’ and ‘Inference’ Chips for Enhanced AI Performance
1 day ago /
Read about 0 minute
Author：小编
On April 23, reports emerged indicating that, on April 22 (U.S. time), Google unveiled its eighth-generation Tensor Processing Unit (TPU) at the Cloud Next '26 conference. For the first time, the company has split AI model training and inference tasks across two specialized chips: the TPU 8t for training and the TPU 8i for inference. These new TPUs are anticipated to officially launch later in 2026. The TPU 8t is tailored specifically for model training, delivering 2.8 times better cost-performance than its predecessor. Meanwhile, the TPU 8i is optimized for inference tasks, offering 80% higher performance and featuring three times the SRAM capacity per chip compared to the previous generation.
-
C114 Communication Network -
Communication Home
7 X 24 Track global technological trends
