Here's what actually issued. On June 21, 2022, NVIDIA Corporation was granted US11367160B2, "Simultaneous compute and graphics scheduling," with a long inventor list including Jack Choquette and Emmett Kilgariff — core GPU-architecture names. The CPC codes are graphics G06T 1/20 plus processor-scheduling classes G06F 9/3851, 9/3867, and 15/8007.
The mechanism is scheduling. A modern GPU is asked to do two very different kinds of work: graphics (rendering pixels) and compute (the matrix math behind neural networks). Running them one after another wastes the chip; running them together, on shared execution units, keeps the hardware busy. The claim covers a method for scheduling both kinds of work simultaneously so the GPU's resources stay saturated.
This is not a machine-learning algorithm — it's the silicon-level plumbing that makes machine learning and graphics coexist efficiently on one device. That's precisely why it's strategically central for NVIDIA: its advantage is not only fast cores but the scheduler and architecture that extract throughput from them. AI workloads increasingly run alongside graphics in the same applications, and owning the co-scheduling method protects a hardware differentiator.
On scope: granted B2, enforceable, but the claims describe a specific simultaneous-scheduling mechanism. They do not cover GPU scheduling in general. The processor-architecture CPCs and claim 1 together define the boundary.
The takeaway: US11367160B2 is a reminder that a lot of NVIDIA's most defensible AI IP isn't an algorithm at all — it's the low-level hardware scheduling that turns transistors into throughput, authored by the architects who design the chips.