Google to make its own custom server chips as well

Get a free Techzine subscription!

Google has announced that it plans to design its own SoCs for use in servers. To assist in the design process, the company has hired Intel veteran Uri Frank.

In a blog post, Google says that Moore’s Law no longer provides great strides in computer speed. To ensure that improvements in speed and efficiency continue, Google wants to focus on developing custom chips.

Different components on a single chip

Google wants to do this by designing its own SoCs. The company wants to move away from the model where the processor, networking, storage, custom accelerators and memory come from many different manufacturers and together make an optimised system. By combining all these chips into a single SoC, Google expects to achieve efficiency gains. In doing so, the company also wants to circumvent bandwidth and latency problems that arise due to the physical distance between different components.

Uri Frank

To help design these SoCs, Google has teamed up with Uri Frank. Frank has nearly 25 years of experience designing processors and spent more than 20 of those years at Intel. There, he even held the title of Senior Director of Engineering between 2016 and 2020.

Arm or x86?

Google has not announced which architecture it wants to base its new chips on. Arm is the obvious choice. With this, the company would follow in the footsteps of AWS with its Graviton processors and several other cloud vendors that offer servers with Arm.

However, it is also possible that Google will go for x86, given Frank’s experience with that architecture. However, it is not possible to license the x86 architecture, so Google would have to turn to Intel or AMD for x86-based SoCs. AMD does offer such chips, such as those found in recent Sony and Microsoft game consoles.

At AWS, the move to self-designed Arm chips seems to be catching on. The company is deploying the technology more widely, while scaling back its use of Intel Xeon chips.