Weekend Listen: Apple’s Former Chief Chip Architect on the AI Chip Race

SPEAKER_1 (0:02)

Bloomberg Audio Studios, Podcasts, Radio, News.

Mark Gurman (0:17)

Welcome to Power On. We talk about Apple and the consumer tech industry. I'm Mark Gurman, and today we have a very interesting conversation and a very special guest, Gerard Williams III, joining us to talk about the AI datacenter boom, what he's building at his new startup, Nuvacore, what he built at Nuvia and Qualcomm, and of course, what he did at Apple as the chief architect on their CPU technology. He worked on chips starting from the A7, the first 64-bit Apple processor, all the way up to the Apple M1 chip. We get into his time there, we get into his thoughts on John Ternus and Johny Srouji, and what it was like when he first came up with the idea, almost two decades ago to build a server-side CPU at Apple. And stay tuned till the end, we have another round of Apple Trivia.

Just to give some background to those listening, Gerard has had an amazing career, and there's probably no one better to talk to right now about the AI data center boom, the silicon industry and all of that coming together.

Gerard, as you may be interested if you're listening to this podcast, he worked at Apple for a long time. Most recently, he was a senior director in charge of CPU work, platform work. And after that, he went to start a company called Nuvia, which made cloud silicon. It was acquired by Qualcomm. And most recently, he left Qualcomm, where he was a senior VP, to start a new company called Nuvacore. So Gerard, thanks for joining the Power On Podcast. Thank you so much for being here. And good to have you, man.

Gerard Williams III (1:57)

Oh, no, it's wonderful to finally get a chance to talk to you one on one.

Mark Gurman (2:01)

Let's continue your journey at Apple. So Gerard Williams III, you show up at Apple in what year?

2010 And so that silicon in the cloud was one of your first ideas. So who hired you? Who brought you on? What was the mission? Was this to develop the A7? Was that the first thing, the first 64-bit Apple chip? Tell me, tell me about getting to Apple, getting started and, and what you did there from the beginning.

Gerard Williams III (2:27)

Yeah, I joined Apple in, I think it was February of 2010 I joined a team called Platform Architecture.

In the early days of Apple, that was a very small engineering team that was focused on literally looking at all silicon and hardware being designed in the company. And I was one of a handful of those engineers that was a part of that team. And we explored every design that would come across the desk of the senior execs and decide with them whether or not that that was a possible thing that could be developed and created in an Apple product. I had the opportunity to become Apple's chief architect of CPU development. The first product that saw the light of day that I worked on was the Apple A7. But, yeah, the engineering team, we focused pretty heavily in silicon development.

Understanding that silicon, developing a roadmap for that silicon, and trying to make sure that the products year on year showed benefit.

Mark Gurman (3:33)

Apple really pushes this narrative about performance per watt on all its silicon. And when the M1 came out, which you worked on, I want to get into that. They were really pushing performance for watt versus Intel. It's not only about the speeds and feeds, but it's about the power efficiency. It's about battery life. Was that something that you were pushing from the beginning? Was that sort of one of your early innovations?

Gerard Williams III (3:54)

I would say that even predates Apple.

Mark Gurman (3:58)

Tell me about that, because you were at TI and ARM as well. I want to hear about the pioneer of this performance per watt idea, because it's the biggest thing on device silicon right now.

Gerard Williams III (4:08)

Even predating Apple, if you go all the way back to during the ARM days when I was there, we used to look at devices and the products that we were shipping. And in those days, back in the, I call it late 90s, early 2000s, you were talking about devices that were sipping 50 milliwatts, 100 milliwatts. So to you as an engineer, expending one milliwatt was huge. It was huge. It was a large percentage of the core itself.

That grew over time, but that philosophy never changed. And in the world of computing, everyday life, us as users, I carry around laptops. We all do. We carry around cell phones, watches, they all have batteries in them. So burning microwatts, milliwatts is fundamental to those products because it basically means at the end of the day, can I make it through the working day with my product doing everything that I want to do and it still survives. And if that's the case, you've succeeded. The beauty of that is once you understand those fundamental properties about how and where power goes and why, you can bring that to the data center. Okay, it's a property that people don't think about. You hear people throw out numbers like gigawatts and multi-gigawatts.