India, Oct. 12 -- Microsoft Corp. (MSFT) has announced its third-generation Maia AI chip, designed to handle AI tasks across its products and services, and is expected to deliver 40% more tokens per watt than leading GPU-based systems.
The chip has been developed to handle the actual AI workloads Microsoft runs. The third generation of its Azure Maia program focuses on AI inference, which is the process of using trained AI models to generate responses and results. It is designed to handle tasks that require large amounts of memory, long inputs and intensive processing.
Maia 300 uses an advanced low-precision format called MXFP4 to reduce the memory bandwidth needed for calculations and storing key-value cache data. This helps AI models pr...