【#Tech24H】Google has unveiled its new custom AI chip - the eighth-generation custom TPU, including the TPU 8t for training and the TPU 8i for inference. This marks the first time Google has separated training and inference chips. The TPU 8t excels at large-scale compute-intensive training, offering higher compute throughput and scalable bandwidth, aiming to shorten model development cycles from months to weeks. The TPU 8i features higher memory bandwidth and is designed for latency-sensitive inference workloads. For the first time, both chips run on Google’s Arm-based Axion CPU hosts to optimize system performance. The TPU 8t and TPU 8i will be available later this year and can be used as part of Google's AI Hypercomputer. [ By Zhang Liyan | Tang Ruohan ]