硬件/软件协同设计
Search documents
英伟达RISC-V,去年出货10亿颗
半导体行业观察· 2025-03-03 01:06
Core Insights - NVIDIA has chosen RISC-V as its embedded microcontroller architecture, marking a significant shift from its previous Falcon processors, with an estimated shipment of around 3 billion Falcon processors over 10 years and a projected shipment of over 1 billion RISC-V processors by 2024 [2][13]. Group 1: Transition to RISC-V - The transition from 32-bit Falcon cores to RISC-V was driven by the need for 64-bit capabilities, with NVIDIA developing various RISC-V cores for different applications, including a dual-issue out-of-order core and a 32-bit version for space-constrained applications [3][4]. - NVIDIA's RISC-V microcontrollers are integrated into their products, with each chip typically containing 10 to 40 RISC-V cores depending on the configuration [2]. Group 2: Custom Extensions and Applications - NVIDIA has developed several custom extensions for RISC-V, enhancing performance and security, including a unique 2kB page size extension that improves legacy software performance by 50% [5][6]. - The Peregrine subsystem, which includes RISC-V cores and various peripherals, allows NVIDIA to flexibly deploy over 30 system control and management applications without the need for independent development for each application [6][9]. Group 3: Software and Security - All applications utilize a single software stack, enabling significant reuse of components such as boot, operating systems, and libraries [7]. - NVIDIA employs an internal security team to identify vulnerabilities, and the Peregrine subsystem's architecture allows for the isolation of different software applications, enhancing security [8][11]. Group 4: Key Reasons for Choosing RISC-V - Customization capabilities allow NVIDIA to optimize chip designs by using the basic ISA as a building block and adding specific extensions [13]. - Hardware/software co-design ensures that hardware is optimized for software workloads, improving efficiency and performance [13]. - RISC-V's configuration options enable NVIDIA to avoid over-specification of standard processors, saving costs and development efforts [13][14].