This week's most important news in the semiconductor market didn't come from the traditional semiconductor companies but from some of the largest semiconductor buyers in the world.

First, Apple announced upgrades for the Macbook Pro lineup, powered by the internally designed M1 Pro and M1 Max SoC based on ARM architecture. Apple first introduced the M1 processor in late 2020 as a replacement for Intel processors for some of the lower-end laptop and desktop products. The latest M1 Pro and M1 Max SoC are far more powerful than their predecessor. M1 Pro uses 10 CPU cores and 16 GPU cores, offers 200GBps of memory bandwidth and delivers double the graphic performance over the M1. Apple claims M1 Pro provides 1.7x the performance and 70% power reduction compared to Intel's latest 8-core laptop processor. The flagship M1 max doubles the GPU cores to 32, doubles the memory bandwidth to 400GBps, and delivers 4x the graphic performance over the original M1 SoC. The M1 Max is comprised of a whopping 57 billion transistors. That is even more than the A100 GPU, the latest AI chip from Nvidia. As a result of the new chips, the new Macbook Pro offers 7 to 10 hours longer battery life than the previous version using Intel processors. Apple once again shows the power of integration and optimization across hardware and software.

Moving to the cloud providers, Alibaba announced its internally designed server processor Yitian 710. Based on ARM v9 architecture and 5nm process node, Yitian 710 has 128 CPU cores with 3.2Ghz of frequency and 60 billion transistors. In the SPECint test, the chip scores an impressive 440, matching some of AMD's most powerful server processors and crushed Intel. There are questions around whether Alibaba designed the chip independently or with some outside help. Dylan Patel at SemiAnalysis offered some fascinating observation that is worth checking out. But regardless, it is likely that Yitian 710 is heavily optimized for Alibaba's own workload.

Another piece of news that is less widely covered, Microsoft posted a job opening, looking for a "Director of SoC Architecture" for its Surface division. As of the writing, the company is no longer accepting applications. The Surface product family has products running on both ARM and x86 processors, but mostly on standard commercial processors without much modification. It is not clear if the software company will go as far as designing an SoC itself, but from the job description, it is clear that optimization across silicon, hardware, and software is the way forward.

• "A fundamental part of our strategy is bringing productivity and mobility together through devices that enable new experiences - helping people and organizations unlock their creativity, passion, and potential. Creating these devices involves a close partnership between silicon, hardware and software engineers, designers, and manufacturing. We are currently building the next generation and Surface needs you!"

On Marvell's analyst day a couple of weeks ago, CEO Matt Murphy stated that we had entered the "era of cloud-optimized silicon" since every cloud has unique requirements for the processors powering them. Optimizing the underlying silicon based on the unique workload will become more prevalent across the industry. Based on that vision, he has positioned Marvell as the extension of the customers' engineering team, to partner and co-invest with the customers. The new developments this week is a clear validation of that vision. 

The era of cloud-optimized silicon

Close

The era of cloud-optimized silicon