Google unveils TPU 8tx and 8i Max with confidential computing
Google introduces TPU 8tx and 8i Max, adding confidential computing plus larger memory and context capacity for both inference and training.
Google has introduced two new TPU chip lines, the 8tx and the 8i Max. The two lines add confidential computing along with larger memory and context capacity, aimed at both inference and model training needs.
Confidential computing is a notable addition at a time when data used for AI is increasingly sensitive. The mechanism aims to protect data and models while they are being processed, rather than only protecting them when data is stored or in transit.
Larger memory and longer context directly affect the kinds of tasks the infrastructure can take on. A larger context allows more documents or more processing steps to be handled at once, while more memory helps reduce the need to split work across multiple runs.
For enterprises, infrastructure options with confidential computing are usually tied to compliance and data control requirements. Weighing cost, inference performance and the level of data protection therefore becomes part of the architecture decision, not just a hardware choice.
Source: blog.google
