Whether you’ve converted (or plan to convert) to an AMD-powered system or not, there’s no denying AMD’s Ryzen series of CPUs injected more life into the overall CPU market in 2017 than we’ve seen in almost a decade. Thanks to a combination of vastly improved thermals, performance, and core counts, AMD has taken the fight to Intel at every price point where the two face off. Intel, in response, has added Hyper-Threading to CPUs that historically don’t have it, increased the core counts it intends to offer at the same price points for its mainstream desktop CPUs (at least, if rumors are accurate), and slashed its own per-core pricing for its top-end HEDT processors. Good times all around.
According to AMD’s desktop CPU marketing manager, Don Woligroski, all this progress took place at what he calls a worst-case scenario for AMD. In a recent interview with WCCFTech, embedded below, Woligroski reflects on the difficulty of bringing Ryzen to market and how it impacted the CPU design:
I’ve said this before and I think it holds true. Zen, Ryzen, was the worst case scenario. It was a brand new architecture on a brand new node. So the worst case scenario we could’ve possibly had, and it’s pretty good. You can get to over 4.GHz [referring to the CPU’s maximum clock speed].
We agree with Woligroski — and we’d go farther. Ryzen wasn’t just a worst-case scenario in terms of hitting a new node at the same time AMD debuted a new architecture — it was a worst-case scenario for a lot of reasons.
AMD presumably knew that Bulldozer would never deliver the performance it hoped for by early 2011, at the very latest. From that point forward, all the company had were bad choices. It could bet on a new, clean-room architecture initiative and hope to survive long enough to bring it to market, or it could bet on repairing Bulldozer. Historically, both AMD and Intel have preferred to evolve existing architectures over launching new ones precisely because new CPU architectures are so risky. It takes 4-5 years to develop and bring a new CPU design to market from scratch, and there’s always the possibility that the assumptions you confidently made in 2011 will no longer make sense when a part hits store shelves.
But breaking the choice down to a binary between fixing or not-fixing Bulldozer still misstates the risk. During this same period (let’s call it Q1 2011 through Q4 2012), AMD also announced it would launch itself into ARM servers, acquired a server vendor (SeaMicro), canceled its Krishna and Wichita CPUs at GlobalFoundries (these became Kabini and Kaveri at TSMC), won the contracts for and began work on the Xbox One and PS4, launched the Radeon HD 7000 series and began work on Hawaii (aka GCN 1.1), committed to improving its FX-derived CPU cores with what resources it could spare, and began work on the CPUs that would become K12 (never launched) and Ryzen (debuted in March 2017).
It did all this while cutting its overall RD investment; the graph below shows Intel, NV, and AMD RD costs through the end of 2015. I don’t want to imply AMD faced no penalties for doing so, because you can see some of the cost in the company’s delayed GPU launches as well as the fact that all of the Ryzen CPUs are cookie-cutter implementations. Then again, AMD still got a core it could scale through its own markets with a cookie-cutter implementation in the first place. That’s something Bulldozer could never do.
Add the huge contraction in PC sales the last six years and AMD’s survival has been far from assured, at multiple points. Ryzen, therefore, wasn’t a worst-case scenario for AMD simply because of its node or architecture. The entire process (pun intended) of bringing Ryzen to market was a worst-case scenario for AMD, nearly without exception. It’s genuinely surprising that Ryzen is as strong as it is, given the headwinds AMD faced in bringing the chip to market.
All of which raises the question: What comes next?
Our cloudy crystal ball suggests 2018 will see the usual set of frequency tweaks and SKU adjustments that are typical of a product on a maturing 14nm process, and we’d expect to see a modest frequency boost within the same TDP and more efficient CPUs hitting lower TDPs further down the stack. AMD has previously implied it would do a 14nm+ spin on Ryzen, even though that terminology is typically used by Intel.
It’s possible that AMD will pull in some architectural changes as well, though we think most of those will likely be reserved for Zen 2, which will arrive on 7nm. GlobalFoundries has said it wants to have 7nm in volume production by the end of 2018, which points to a 2019 launch window for Zen 2. (AMD’s graph above shows a “Zen 3” in 2020, but this doesn’t square with a Zen 2 that debuts in 2019, and GF has publicly said it won’t start High Volume Manufacturing on 7nm until 2H 2018 at the earliest.)
Intel is expected to counter Zen with improved core counts and single-thread performance as part of its upcoming Coffee Lake family of CPUs. But AMD has enough leeway in its current market position to retrench itself without gutting its prices the way it did during the long, sad slide of the Piledriver years. It’s going to be an interesting 12-18 months, even if Zen 2 hits towards the end of that timeframe.
Now read: How L1 and L2 CPU Caches Work, and Why They’re an Essential Part of Modern Chips