New AMD Patents

AMD's 10 newest filings sort into three clusters:

• Multi-chip memory sharing. Two filings cover interconnected processor layouts where chips share memory with neighbors, including a shared memory block that sits between AI processor groups and a grid where adjacent chips access each other's memory directly.
• GPU efficiency and scheduling. Four filings target how GPUs manage their own workload, covering self-directed math scheduling without CPU involvement, keeping GPUs busy while memory catches up, trimming memory waste in ray-traced games, and redirecting unused storage toward AI processing.
• Graphics and game image quality. The remaining four filings address visual output, including AI-driven texture compression, feeding AI training examples to sharpen game images, drawing less detail during camera movement, and displaying visuals before a PC finishes booting.