Google introduces two new TPU chips for private AI and bigger jobs

In short: Google has announced two new AI-focused chips, TPU 8tx and TPU 8i Max, to support more private AI work and larger AI tasks.

What happened

Google said it is adding two new versions of its Tensor Processing Units, or TPUs, which are chips it designs for AI work (think of them as specialized engines built to run AI faster). The company said the new chips help it rely less on chips from other companies in its cloud business.

The first chip is called TPU 8tx. Google said it was designed with AMD and is aimed at “confidential” training and use of AI models. In simple terms, it is built to protect sensitive data while it is being used, not just when it is stored.

Google also said TPU 8tx can do some “reasoning” work while it trains a model. That could help a company train or adjust a smaller AI model using its own private information without moving that data between different devices.

The second chip is called TPU 8i Max. Google said it combines two of its earlier TPU 8i chips into one package, which doubles memory and processing power. More memory can help an AI system handle a larger amount of information when it answers a question.

Why it matters

AI systems are expensive to run, and the chips that power them are in high demand. By building more of its own chips, Google can control costs and supply for its cloud services. For customers, the biggest practical benefit may be more options for keeping sensitive business data protected while using AI, plus the ability to run larger AI workloads without hitting memory limits.

Source: TechCrunch AI