Google Cloud推出TPU 8tx與TPU 8i Max,分別因應不同AI運算需求
Google Cloud introduces TPU 8tx and TPU 8i Max for different AI workloads
Updated at: November 10, 2026 at 03:55 AM
Google Cloud於11月9日推出兩套執行人工智慧模型的系統。TPU 8tx旨在保護處理中的敏感資料,TPU 8i Max則面向需要同時處理大量資訊的模型。
Google Cloud introduced two systems for running artificial intelligence models on November 9. The TPU 8tx is designed to keep sensitive information protected during processing. The TPU 8i Max is aimed at models that need to handle large amounts of information at once.
兩套系統都屬於Google第八代張量處理器。該公司4月曾公布用於大規模模型訓練的TPU 8t,以及用於需要快速回應的推理任務的TPU 8i。
Both systems belong to Google’s eighth generation of Tensor Processing Units. In April the company announced TPU 8t for large-scale model training and TPU 8i for reasoning tasks that require quick responses.
8tx使用與8t相同的基礎晶片。Google表示,新系統結合了可在運作期間保護資料及模型權重的硬體。
The 8tx uses the same underlying chip as the 8t. Google said the new system combines that chip with hardware that keeps data and model weights protected even while the system is using them.
Google把8i Max定位為解決記憶體瓶頸的方案。如果記憶體供應資訊的速度跟不上,處理器就可能得停下來等候。
Google describes the 8i Max as its answer to a memory bottleneck. A processor can end up waiting for information if the memory system cannot deliver it quickly enough.
Google表示,8i Max的運算能力、晶片內SRAM及高頻寬記憶體的頻寬都是8i的兩倍。同樣大小的機架可容納兩倍的晶片。
Compared with the 8i, Google says the 8i Max doubles computing capacity, on-chip SRAM and the bandwidth of its high-bandwidth memory. It fits twice the silicon into the same rack footprint.
公布這兩套系統後,第八代TPU已有四套。Google表示,這是它首次在同一年推出四套TPU系統。
The announcement brings the eighth generation to four systems. Google says this is the first time it has introduced four TPU systems in one year.
