Nvidia has integrated dozens of RISC-V cores into its modern GPU architectures. Depending on the specific model, each graphics card contains between 10 and 40 of these specialized processing units. They operate independently of shader calculations, handling auxiliary tasks instead.
These cores are locked for user programming, and any unauthorized modification prevents the GPU firmware from functioning correctly. One particular core has been responsible for executing significant portions of the graphics driver logic since 2018, marking a shift in internal management.
Previously, Nvidia relied on its proprietary Falcon architecture, known as the Fast Logic Controller. This design utilized a pipeline structure with variable-length instruction sets, focusing on minimizing chip area and managing memory latency rather than raw performance.
The company eventually abandoned the Falcon architecture due to its limitations. Specifically, the design lacked a 64-bit addressing space and could not support the execution of a full operating system, prompting the transition to the RISC-V standard.
