Google hires Intel chip designer to create its own server socs
Google is setting up a team that will design its own socs for servers. Google has appointed Uri Frank as its vice president of engineering. Frank worked at Intel for more than twenty years, including on the Core processors.
Google wants to focus on integrating as many parts as possible into one chip. Until now, the search giant has mainly focused on making components that go on motherboards, such as the tensor processing units and the video processing units, which Google has been making since 2015 and 2018 respectively.
According to Google, making components for motherboards is no longer enough. To enable better performance with lower consumption, further integration of hardware is necessary. That is why Google wants to make socs that combine various components. “The soc is the new motherboard,” writes the search giant.
Google wants to build its socs with components from different manufacturers. “We buy where it makes sense and build it ourselves if we have to,” the company writes. Google lists a number of examples of parts that should be included in the server socs. In addition to the cpu and tpu, these are components for video transcoding, encryption, compression, remote communication, data security ‘and more’. In its own words, Google wants to create ecosystems that benefit ‘the entire industry’. It is not yet known on which CPU architecture Google wants to base its socs.
Uri Frank has been appointed as the top man of the team that will design the server socs. That will be done by a new team in Israel. Frank writes on his LinkedIn page that vacancies will be posted soon. According to the Israeli website Calcalistech, Google wants to hire ‘hundreds’ of chip designers. Since the team is now being set up, it could be years before Google actually releases its own server socs.
In making its own server socs, Google follows Amazon’s strategy, which has been making its own Graviton socs based on the Arm architecture for several years now for its AWS platform. Microsoft is also rumored to be working on its own Arm chips for its Azure cloud services.
Google has been making its own tensor processing units for deep learning since 2015