Modern GPUs are highly complex system-on-chips with a massive number of functions and resources — starting from compute resources and power management and all the way to display engines and
security — that must be managed.
These things are now managed by 10 to 40 custom RISC-V cores developed by Nvidia, depending on chip complexity. Nvidia started to replace its proprietary microcontrollers with RISC-V-based microcontroller cores in 2015, and by now, virtually all of its MCU cores are RISC-V-based, according to an Nvidia slide demonstrated at the RISC-V Summit.
By now, Nvidia has developed at least three RISC-V microcontroller cores: NV-RISCV32 (RV32I-MU, in-order single-issue core), NV-RISCV64 (RV64I-MSU, out-of-order dual-issue core), and NV-RVV (RV32I-MU, NVRISCV32 + 1024-bit vector extension). These cores (and perhaps others) replaced the proprietary Falcon microcontroller unit based on a different instruction set architecture. In addition, Nvidia has developed 20+ custom RISC-V extensions for extra performance, functionality, and security.
Perhaps the most important RISC-V-based part of Nvidia GPUs is its embedded GPU System Processor (GSP). According to Nvidia's website, the first GPUs to use RISC-V-based GSP were based on the Turing architecture. This GSP offloads Kernel Driver functions, reduces GPU MIMO exposure to the CPU, and manages how the GPU is used.
Since MCU cores are universal, they can be used across Nvidia's products. As a result, in 2024, Nvidia is expected to ship around a billion RISC-V cores built into its GPUs, CPUs, SoCs, and other products, according to one of the demonstrated slights, which highlights the ubiquity of custom RISC-V cores in Nvidia's hardware.