Given how much AMD dominates the world of CPUs and gaming in general, it might seem strange to picture such a company on the verge of bankruptcy, being billions of dollars in debt and with few good products on offer. That was the state of AMD over 20 years ago, but thanks to some inspired engineering decisions, those days are now in the distant past. And in a recent interview, CEO Lisa Su tells us that one such choice was betting almost everything on chiplets.

The interview took place on Reid Hoffman’s YouTube channel, a venture capitalist and co-founder of LinkedIn, amongst many other strings to his bow. The two hosts chatted with AMD’s CEO, Dr Lisa Su, about her past and what led her to become a chip engineer, as well as all things AI, of course.

For me, the most interesting part was when Su talked about AMD’s decision-making process, which ultimately led to the likes of the Ryzen 7 9800X3D being the most sought-after CPU by PC gamers.

“We were in a place where the world in semiconductors, for 30 years, really was driven by Moore’s law—the idea that every two years or every 18 months you could double the productivity—and so it was all about scaling, scaling, scaling, scaling. But scaling was changing and we saw that change, and yes, we made a few bets.”

When it comes to designing any kind of HPC (high performance computing) processor, any bet that you make takes time to bear fruit, if at all. I’m not talking about a period of a few months, or even a year or two. The timescales involved are around a decade or so, and while that might seem like a ludicrously long time, in the world of processors, that only covers a handful of product cycles.

In other words, such bets carry considerable risk. But one in particular turned out to be more than worth it.

Su continued: “We made the bet that scaling would slow down, and we had to innovate in different ways, and perhaps one of the largest innovations is we were the first to really put high performance computing in sort of breaking it up into what we would call chiplets.”

It’s hard to overstate just how important this move was for AMD and perhaps the CPU industry as a whole. Its Zen architecture, launched eight years ago in 2017, was considerably better than earlier designs but still relied on monolithic fabrication (i.e. everything in a single die), which can be costly to scale across every sector.

Splitting its CPUs up into multiple dies, aka chiplets, meant that AMD only needed to make three discrete chips, but it could then use them to make processors for laptops, desktop PCs, workstations, and full-blown servers. Intel, on the other hand, was making completely different chips for each market.

Image 1 of 3

(Image credit: Fritzchens Fritz)Image 2 of 3

(Image credit: AMD)Image 3 of 3

(Image credit: AMD)

It might seem like such an obvious move these days, but given how earlier mistakes backfired drastically for AMD, you’d be forgiven for thinking that its leaders were cautious about it all. Not so, it would seem.

“You know those meetings, back in the day, when I would sit with our technical fellows and we would make that big decision of ‘hey, are we going to bet the company’s roadmap on chiplets?’ and we said yes,” said Su, looking pleased as punch. “We had a lot of learning. I mean, I can say there were products that worked really, really well, and there were products that worked not so well, but we learned a ton from the process of that.”

Another interesting topic that’s discussed in the interview is how AMD has focused almost entirely on HPC. For example, despite the breadth of its processor portfolio, it doesn’t make chips for phones. Dr Su explains why this is:

“Our driving force at the time was to decide what we could be best at. I really believed—my CTO Mark Papermaster has been my partner on this journey—we believed that we could be bet the best in high performance computing, and at the time, high performance computing wasn’t as exciting as it is today because AI wasn’t as big as it is today and the cloud was you know kind of on a different path.”

(Image credit: AMD)

More specifically, she remarks: “There was a lot of conversation on ‘hey Lisa, you know AMD should be making smartphone chips’ and then when you look at where things have evolved, you actually realise that companies have sort of fundamental DNA in terms of what they’re good at. We are just really good at building big computers, that’s what we think about every single day.”

It’s certainly all worked out very well for AMD. It makes the best CPUs, it dominates the console sector, it has a significant foot in the FPGA market, and as things currently stand, it boasts share prices more than 4.5 times higher than Intel. Despite its vast AI revenue stream, Nvidia’s shares are just 10% higher than AMD’s.

With everything Su says in the interview, it’s clear what the next decade of AMD will look like: more chiplets, more HPC, more being the best at something. A stark contrast to 20 years ago.

From PCGamer latest via this RSS feed