MobileNews

Report: Google custom silicon work continues with ARM server chips

Following “Argos” VCU chips that help YouTube process videos more efficiently and Tensor, Google is working on custom server chips.


9to5Google has a rebooted newsletter that highlights the biggest Google stories with added commentary and other tidbits. Sign up here!


According to The Information, there are two 5nm, ARM-based server chips in development. Maple is “based on an existing design from chipmaker Marvell Technology,” while Cypress is an “in-house design developed by a team in Israel.” 

Both come from the server chip design team led by a 25-year veteran of custom CPU design and delivery that previously worked at Intel. Uri Frank became Google’s VP of Engineering for server chip design in March of 2021.

The company previously said that it was focused on Systems on Chip (SoC) where “multiple functions sit on the same chip, or on multiple chips inside one package” for “orders of magnitude better” latency and bandwidth between components to greatly reduce power and cost. It follows Google’s work on Tensor Processing Units (TPUs) to speed up ML workloads, custom SSDs, network switches, and network interface cards.

Google Cloud is said to be targeting performance that’s “at least comparable performance to that of Intel’s and AMD’s server chips.” 

That way, Google could avoid buying those vendors’ chips at retail price and instead pay to produce its own custom-made chips at cost, said one of the people. Its ultimate goal is to improve its chips over time so they offer 20% to 40% better price performance than Intel’s equivalents, the person said.

The design for Maple was just completed and handed over to TSMC (Taiwan Semiconductor) for trial production. (Samsung is used for Tensor on phones.) It’s considered a backup to Cypress, which is expected to meet the same milestone in Q2. According to The Information, mass production of Google’s custom server chips is expected for the second half of 2024, with data center deployment “as early as 2025.”

More on Google Cloud:



Author: Abner Li
Source: 9TO5Google

Related posts
AI & RoboticsNews

Nvidia and DataStax just made generative AI smarter and leaner — here’s how

AI & RoboticsNews

OpenAI opens up its most powerful model, o1, to third-party developers

AI & RoboticsNews

UAE’s Falcon 3 challenges open-source leaders amid surging demand for small AI models

DefenseNews

Army, Navy conduct key hypersonic missile test

Sign up for our Newsletter and
stay informed!