Grace Architecture

The Grace Architecture is the design of a CPU (central processing unit) created by NVIDIA. Just as car makers each design engines a bit differently, chipmakers design CPUs differently too—Grace is the name of NVIDIA's own approach to that design.

What makes this CPU stand out is that it doesn't work alone. In chips like RTX Spark, a Grace Architecture-based CPU is combined into a single package together with a GPU handling graphics processing and memory. Compared to attaching separate components one by one, the shorter distances between them allow devices to be thinner and quieter, while still delivering enough power to handle heavy AI tasks locally, without needing an internet connection.

This approach of combining CPU, GPU, and memory into a single chip is similar in concept to Apple's own chips used in Macs. It serves as a core component in NVIDIA's strategy for local AI hardware in the laptop and mini PC markets.

Articles explain that "the Grace Architecture-based CPU includes up to 20 cores, while the Blackwell RTX GPU includes up to 6,144 cores." One point that can cause confusion here is that Grace isn't the name of the GPU handling graphics. Graphics are handled by a separately named GPU called Blackwell, while Grace is the name on the CPU side, responsible for computation and control alongside it.