Analysis Here’s something that should raise eyebrows in the data center world: Facebook’s parent company Meta has hired a veteran network chip engineer from Intel to lead silicon design efforts in the engineering group. infrastructure hardware from the Internet giant.
Jon Dama started as Director of Silicon in May for Meta’s Infrastructure Hardware Group, a role that makes him “responsible for multiple design teams that innovate the data center at scale,” according to his LinkedIn profile. . In a blurb, Dama said a team is already in place at Meta, and he hopes to “scale the next data processing doublings” with them.
Although we couldn’t confirm this, we think it’s likely that Dama reports to Alexis Bjorlin, vice president of infrastructure hardware at Meta, who previously worked with Dama when she was general manager of the Connectivity group at Intel before spending two years at Broadcom.
Unfortunately, Dama hasn’t revealed more about his role beyond basic details, though we suspect Meta likely leveraged his expertise to meet some of the internet giant’s biggest data center needs. , including bolstering its recommendation engines and hype-filled “metaverse.” ambitions.
Dama helped lead the development of Intel’s IPU with Google
At Intel, Dama helped lead silicon efforts for a variety of networking hardware for the data center and telecommunications markets. This includes the giant’s nascent x86 Infrastructure Processing Units (IPUs), which belong to a growing class of chips, also known as SmartNICs and Data Processing Units, that accelerate input/output intensive workloads, offloaded from host processor cores, for networking, storage and security.
One piece of information that Meta probably found appealing about Dama is that it helped guide Intel’s co-development with Google on the Mount Evans IPU ASIC, which aims to provide maximum performance in so-called hyperscale data centers. During his two years of involvement in this field, his title was Director of Cloud and Silicon IPU in the CTO Office of Intel’s Connectivity Group.
Most recently, Dama served as Executive Director of Silicon Engineering for the Connectivity Group. It was in this role that Dama was responsible for the “full silicon development lifecycle” for the group’s products, which now reports to Intel’s Network and Edge Group business unit.
In this role, Dama spent significant time on the Tofino 2 and Tofino 3 ASIC switches, which were derived from Intel’s 2019 acquisition of Barefoot Networks. He also led the development of “the new programmable pipeline logic and IP Ethernet” with a focus on improving energy efficiency.
Another potential point of interest for Meta is how Dama helped triple the size of its silicon engineering team within Intel’s Connectivity Group, according to its LinkedIn page. As group leader, Dama said he fostered a “culture of excellence”, which allowed the team to develop two projects in parallel and move to a “customer owned tooling” model.
Dama spent more than 10 years at Intel, which he joined when it acquired network ASIC maker Fulcrum Microsystems in 2011.
Why Meta could use Dama’s expertise
So that brings us back to Meta. Why did the internet giant hire a veteran silicon engineer whose specialty revolves around various network chip technologies?
We don’t expect Meta to explain the implications of Dama’s role, although the web titan has been a bit open over the past few months about how it responds to changing hardware requirements. There have also been reports detailing Meta’s recent custom silicon efforts.
Perhaps most telling is a December 2021 article by industry analyst Patrick Moorhead, who interviewed Meta executive Jason Taylor about why the company hired Bjorlin to lead its hardware group. infrastructure.
At the time, Taylor said that while the main focus of Meta’s silicon strategy was to work with Intel, Nvidia, AMD and Broadcom, the company was exploring creating custom silicon for things like memory and network bandwidth to improve performance. recommendation engines, which underpin other Facebook and Meta properties.
Moorhead wrote in his article that Meta tapped Bjorlin to lead infrastructure hardware to “meet its rapidly growing data center needs,” which includes using a mix of silicon general purpose processors from partners as well as “internal special purpose processors for custom workloads”. “
Chipmakers aren’t quite on the hardware of metaverse devices yet
We have to wonder to what extent Meta plans to increase its reliance on custom silicon. After all, The Information reported last month that Meta was seeking to “control key technologies and reduce its reliance on off-the-shelf silicon vendors.” And while in-house silicon development can be expensive, it can help reduce costs over time, if all goes well.
Another report from several months ago indicated that Meta was developing custom server chips, one to improve the machine learning performance of its recommendation engines and another to improve video transcoding performance for things like streaming live.
We have to imagine that Meta is also thinking about how the right mix of silicon can help it meet the infrastructure needs for its lofty vision of the Metaverse. On the device side, the company has reportedly struggled to create custom chips for new augmented and virtual reality products, leading it to look to Qualcomm for smart glasses.
In a blog post earlier this year, Meta gave us an overview of the infrastructure requirements required by metaverse applications, saying that these applications “will require significant advances in network latency, symmetric bandwidth and overall network speed”. Interestingly, Meta said such work will require “industry-wide collaboration.”
So we have some ambiguity about the extent to which Meta plans to lean towards custom silicon versus silicon from other companies in the future. It’s likely to use a mix of its own chips to speed up some crucial tasks while relying on traditional components for general computing, storage, and more.
We know that Dama already has experience working on silicon that addresses the specific needs of hyperscalers, which could play into new local silicon efforts for Meta or co-development projects like Intel’s with Google. ®