Google's eighth-generation TPUs split training and inference into two chips
Source
Google AI
Author
Google AI
Date
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
Why it matters
Google is splitting silicon between training and AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → serving, and an 80% gain in performance per dollar on the inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → chip feeds directly into what running agents in production costs.
