Intel has been in the news a great deal of late, but the stories themselves haven’t given us much opportunity to talk about what’s happening at the company in a more cohesive way. In many ways, it’s a company at a crossroads, grappling with manufacturing issues at a difficult moment, just as the microprocessor industry is evolving in new and interesting ways.
The following story wasn’t sparked by any single recent event or product launch. Instead, it’s a synthesis of questions I’ve seen authors, pundits, and readers discussing in various articles and forums across the ‘Net. It’s an attempt to capture at least some of the major questions we talk about these days when we talk about Intel, and what those questions say about the company’s near-term business prospects and future plans.
Does Intel Still Build the Best CPUs?
But Intel hasn’t had much luck delivering significant performance gains on a core-for-core basis since Sandy Bridge launched in 2011. Improvements have come incrementally and often relied on SIMD instruction set optimization or specific test cases. So long as Intel faced no meaningful market competition, there was no real argument against the idea that this was simply the new reality of microprocessor performance improvements. Should AMD’s Ryzen 2 prove more efficient and faster overall than the Intel CPU it competes against in 2019 – 2020, however, Intel’s de facto position as the arbiter of CPU performance could be shaken.
Meanwhile, Intel must also contend with the incipient ARM invasion in low-end laptop markets. Here, the nature of the threat is very different and there’s a significant ecosystem advantage for x86 compared with low-end Snapdragon 835 or Snapdragon 850 laptops. Intel doesn’t necessarily need to compete with these systems in terms of battery life, but it needs to make certain its lower-end systems with reduced battery runtime compensate for it with markedly better performance compared with their new ARM competitors.
Neither of these questions is going to be settled in 2018. Second-generation Ryzen won’t hit the market until 2019 and Snapdragon 850 won’t ship until the tail end of this year. No matter how well devices sell in the 2018 holiday season we won’t know anything about their staying power until the end of next year. But rumors that Apple could ditch x86 CPUs by 2020 in favor of their own custom ARM cores have also fueled a discussion that Intel is under fire in all market segments.
Intel’s CPU business also faces a threat from Nvidia, which has dominated the HPC space and won huge new markets for its graphics solutions there and in AI. While Intel can still sell silicon into these GPU systems, it’s selling one core to as many as 8-16 GPUs — and that’s not a ratio the company favors. Intel’s decision to launch itself into the GPU market isn’t just a consumer play, it’s an effort to move into these new, GPU-centric markets.
Does Intel Still Lead the World in Process Technology?
Intel has led the semiconductor industry in new node deployments and process technology for many years, but the company started to put particular emphasis on that leadership as it prepared to battle ARM in the mobile market in 2011 – 2012. At the time, Intel enjoyed a full node advantage over its rivals and was the first company to deploy FinFETs. As it entered the mobile market, Intel was happy to talk up the advantage it would gain from superior manufacturing technology. Those advantages mostly didn’t materialize.
We’ve discussed the economics behind Intel’s failure to win mobile market share long-term in some detail in previous articles (linked below) and won’t revisit the topic here. But suffice to say that Intel’s process advantages did not prove to be the linchpin of its attempts to win tablet or phone share. Meanwhile, 14nm slipped badly, and the 10nm ramp has completely collapsed.
Based on what we know today, Intel has ceded its overall process leadership position by delaying 10nm until the end of 2019. We don’t know how well its 10nm will compare with 7nm from GlobalFoundries, Samsung, and TSMC and until samples of each are in-hand, it’s premature to speculate. But what does seem certain at this point is that Intel’s 10nm will be competing against rival nodes from the three pure-play foundries, rather than leading the pack. That’s scarcely a killing blow to Intel’s semiconductor business, but it’s a far cry from where Intel was five years ago.
Does It Matter If Intel Doesn’t Lead on Process Node?
There’s yet another question buried in these discussions — exactly how much does process node leadership matter in an era where each node’s improvements are smaller than the one before? This point has ramifications beyond Intel. Only certain kinds of silicon designs benefit from moving to FinFETs in the first place, which is why GlobalFoundries has created its 22FDX and 12FDX FD-SOI offerings. EUV will need to be cost-effective before some designs that could benefit from smaller geometries can afford to move to the new node. And even before Intel had to delay its 10nm node, the company had made it clear that 14nm would be a more advantageous manufacturing node for desktop chips until second-generation 10nm was ready.
We don’t know how the delays to 10nm will impact this slide — Intel will be long past 14nm++ by the time its 10nm is ready — but the point stands. Even before 10nm, we knew that cutting-edge nodes weren’t necessarily best for all types of silicon. When the NAND industry switched to 3D NAND, manufacturers like Samsung fell back to older process nodes and built vertically stacked DRAM on a 40nm process. We’re not suggesting Intel has anything like this kind of option, but as the reliability of node shrinks has fallen and alternative processor geometries and chip-stacking methods have become possible, options have proliferated. And that means conventional metrics of success may not mean what they used to.
Given that Intel doesn’t compete in most of the same markets as the pure-play foundries do, it’s not clear how much the 10nm delay practically hurts the company, and the impact may be different depending on which competitor we’re discussing. It may not hurt Intel much at all against AMD, which has focused its competitive efforts in desktops and servers but could hurt it more against ARM, which is attacking from the low-power space.
What’s the Value of Intel’s IDM Status?
Intel is the only remaining integrated device manufacturer (IDM) with a cutting-edge foundry business dedicated almost exclusively to its own products. From 2011 – 2015, Intel portrayed its unique status as a robust positive. Because it designs and builds its own CPUs, Intel can theoretically take advantage of much tighter working relationships than exist between any pure-play foundry like TSMC and its various customers. Intel CPUs are designed specifically for Intel process nodes, using a restricted, Intel-specific set of rules intended to maximize performance and energy efficiency on a given Intel node. This, Intel once implied, would be a critical advantage in mobile.
But those advantages, however real they might be in other markets, never really seemed to materialize in mobile. The first iteration of Core M improved on Intel’s overall power efficiency but poor initial yields and mediocre performance harmed the launch overall and Intel appears to have dropped the Core M branding after Kaby Lake. Intel has evolved its 14nm process node considerably over the past few years, but its custom foundry business has been practically moribund. The company bought the one major customer it managed to attract and the status of its LG deal for 10nm SoC production is unknown. It hasn’t launched a fundamentally new CPU architecture since Skylake.
In short, Intel’s restrictive design rules and high degree of process customization relative to its own parts no longer seems like an obvious, immediate advantage for its own hardware and the company’s lack of customer announcements implies that third parties either have not seen the value in the offering or have been unable to come to other agreeable terms with the company. The repeated delays at 14nm and 10nm have raised questions about how much real, practical advantage Intel currently receives from an arrangement it billed as responsible for its past and future success just a few years ago.
Is Intel Well-Positioned for the Future?
Intel has been on a tear lately in multiple markets. It’s continued to unveil new IoT and AI products for low-power markets, like the Nervana microprocessor, and it wants a major piece of the 5G space. It’s building its own GPU architecture and while it canceled Knights Hill, its HPC-focused many-integrated-core (MIC) processor, it intends to return to this space with a new architecture. It’s announced shipments of Xeons with integrated FPGAs, a feature four years in the making. Former CEO Brian Krzanich made it clear that he intended to focus on building Intel’s market share in multiple emerging spaces as well as doubling down on servers and cloud computing.
I’m not trying to answer as open-ended a question as for whether Intel has hit the right markets in the right fashion to guarantee its own market dominance 4-5 years from now. What we can say is that Intel has clearly made multiple moves to position itself to compete in new and emerging markets. It’s demonstrated strong growth year-on-year in the data center for multiple years running and seems to be an early player in 5G as well. Certainly, Intel seems to be better positioned for 5G than it was to come from behind in the mobile market several years ago.
Other bets, like the company’s move to build its own GPUs, are impossible to predict at this stage. Intel could also be working on a new microprocessor architecture — a true follow-up to Skylake that would be more than another incremental evolution of the same CPU family.
It’s difficult to know what to make of Intel these days. The company faces technological delays that could prove a mild drag on profits or a real competitive disadvantage. It’s moved to enter multiple emerging markets and is in different stages of doing so depending on where you look.
You can make a coherent argument that the 10nm delay won’t be a material problem, ARM will likely be held off by the intrinsic strength of the x86 ecosystem (anybody miss Windows RT? Anybody?), AMD won’t leapfrog Intel on absolute performance, physics will keep clock speed improvements from appearing, and that Intel is already well-positioned with chips like Nervana and in 5G, with exciting new products like FPGA-equipped Xeons and a focus on driving data center engagement that insulates it from the vagaries of the consumer market. Sure, the company has to build a new Xeon Phi microarchitecture and it faces an uphill slog in GPUs, but these aren’t critical flaws, and they aren’t problems different from those Intel has solved before.
Conversely, you can also argue that any disadvantage from the 10nm delay only compounds Intel’s problems vis-à-vis ARM and AMD, that the company no longer can be trusted to have found the best method of building a high performance microarchitecture, that Intel’s Knights Hill cancellation will leave the company out of the market at a key moment for HPC, and that its success with Nervana and 5G won’t mean much if it can’t offer a comprehensive top-to-bottom solution for everything from IoT to HPC-scale AI and ML deployments.
Which of these arguments is truer? I tend to come down somewhere in the middle. I think the impact of Intel’s 10nm delay could be more nuanced than damning. It’s hard to imagine that the 10nm delay would be as large an advantage for AMD or ARM as Intel’s 90nm node and accompanying Prescott processor, which left Chipzilla in the weakest position against its competition than it’s occupied before or since. Intel has clearly made moves to enter some markets early while canceling Knights Hill could hurt it in HPC deployments and its own GPUs are years from coming to market. The opportunities for AMD and ARM to take market share are real, but I wouldn’t go so far as to bet on Intel going down for the count based on these headwinds.