Meta’s first in-house RISC-V-based silicon for AI acceleration is a 7nm component with an operational frequency of 800MHz.
The RISC-V architecture is gaining significant momentum in the tech industry, with Meta announcing its plans to extensively develop hardware and products using this technology. RISC-V, an open-source Instruction Set Architecture (ISA), is an alternative to the x86 architecture predominantly used by Intel and AMD. Its appeal lies in its accessibility and cost-effectiveness, as it eliminates the need for licensing fees associated with using CPU designs.
This shift towards RISC-V is particularly beneficial for Meta’s focus on AI, as the architecture offers power efficiency, performance, low latency, and versatility for various workloads. The senior director of engineering at Meta, speaking at the RISC-V Summit, highlighted the company’s commitment to this transition, which has been in the planning stages for four years.
Meta has already implemented RISC-V in its video transcoding hardware, the Meta Scalable Video Processor (MSVP), which has replaced 85% of the CPUs previously used. The MSVP is operational and manages video uploads on several social media platforms, including Facebook, Messenger, and Instagram.
Meta is exploring RISC-V for developing chips aimed at AI training and inference. This move comes at a time when the tech industry is heavily reliant on Nvidia’s GPUs, such as the A100, H100, and H200, which are important in the advancement of generative AI. However, Meta is choosing to bypass these stages and focus on RISC-V for building AI processors.
Meta’s first in-house RISC-V-based silicon for AI acceleration is a 7nm component with an operational frequency of 800MHz. It includes 128MB of on-chip memory and supports up to 128GB of LPDDR5 RAM. These components are currently dedicated to accelerating AI models in both the training and inference phases, signifying Meta’s significant investment in RISC-V technology for AI applications.
