Op-ed: The desktop CPU isn’t dead, it just needs a swift kick in the butt

Our review of Intel’s new flagship consumer desktop CPU, the Kaby Lake-based Core i7-7700K, was less-than-favorable. Out of the box, the chip runs faster than the i7-6700K that preceded it, but that’s just because it ships at a higher default clock speed. When running at the same clock speed—something easily achievable because these chips are specifically intended for overclocking—CPU and GPU performance is identical.

This feeds into a growing perception that, after several years of modest-at-best performance improvements, Intel is having trouble making its processors faster. Granted, that's not a problem for many casual-to-moderate PC users, and it's not like Intel's chips haven't improved in other big ways in the last half-decade. Power consumption is down, battery life is up, and integrated graphics performance isn't nearly as laughable as it was ten years ago. But for high-end pro users who don't want to spend $1,000 or more on a processor, the lack of performance improvements in Intel's mainstream quad-core desktop processors in particular has been frustrating.

This is an unfortunate reality brought on by the difficulties Intel is having switching to new process architectures. Moving from the 22nm process to the 14nm process in 2013 and 2014 caused several delays that pushed back the launch of the Broadwell architecture and protracted its rollout. The move from 14nm to 10nm is proving even more difficult, breaking Intel's longstanding “tick-tock” development model in which it changed manufacturing processes every two years. If leaked roadmaps are correct, “Cannonlake” laptop chips may move to the 10nm process at the tail end of 2017. But the “Coffee Lake” desktop chips will remain on the 14nm process until well into 2018.

If you don’t know why this is a big deal, here’s a brief explainer: a manufacturing process shrink makes it possible to fit more transistors in a smaller amount of space and reduce the amount of electrical current needed to switch transistors on and off (the type of material used and the type of transistor, two other things that Intel and others often change when they change processes, affects power usage too). This lets chip companies add more features and improve performance (usually by adding more transistors) while keeping power consumption at around the same level as before (or sometimes reducing it). Without process shrinks, improving performance without also hurting battery life becomes that much more difficult.

That's Intel's current predicament. But it doesn’t mean that desktop performance needs to stop improving. It just means that Intel needs to do things a little differently if it wants to make those improvements.

Adding more cores

The most obvious way to improve performance, at least in desktops, is to add more cores. This was what both Intel and AMD started doing a little over a decade ago when improving performance primarily through clock-speed increases began to become more difficult. And desktops have more room for cooling fans, which means that there’s headroom to push their maximum power consumption up without straining systems too badly; a mainstream quad-core desktop processor from Intel today has a TDP of 65W where a mainstream desktop Pentium 4 chip had a TDP of 115W. We wouldn’t want to return to 2004’s power consumption levels, but there’s some wiggle room.

In fact, Intel already does this in its small enthusiast processor lineup. Six, eight, and ten-core processors with a 140W TDP already exist, and multi-core performance is excellent. But these CPUs are ridiculously expensive—you could build a respectable quad-core gaming computer for the cost of the $1000-ish eight-core chip, to say nothing of the $1700 ten-core version. They also use more expensive motherboards with chipsets that are updated less regularly, and they usually trail Intel’s flagship processor architectures by 12 to 18 months. Finally, they need more robust quad-channel memory controllers to keep all of those cores fed compared to the dual-channel controllers in mainstream processors. All those controllers add even more transistors and complexity.

Even with those downsides, adding more cores is probably the easiest and best way to boost performance at the high-end and convince consumers to replace that three-to-five-year-old PC. Enough apps are built to take advantage of multiple cores that users would see benefits immediately. Intel’s Turbo Boost tech (particularly Turbo Boost 3.0, introduced in the aforementioned Broadwell-E CPUs but not available anywhere else yet) can maintain good performance for single-threaded or lightly threaded tasks.

If the leaked roadmaps we cited before are to be believed, Intel may be planning to do this when the “Coffee Lake” processors are released in 2018. The highest-end versions of these CPUs on the roadmap are six-core part, giving users a reason to be optimistic about Intel’s fourth (!) CPU architecture shipped using some version of its 14nm process.

Still, it’s disappointing that Intel has taken this long to improve its mainstream desktop processor performance, and the company could still decide to price those high-end chips far above the prices that current quad-core processors command. Luckily, some serious competition at the high end could both speed Intel up and force its prices down.

Competitive pressure from AMD... and from Intel?

Let’s begin by saying that no one in the last decade has lost money by underestimating AMD. The company’s “Bulldozer” architecture and its many descendants could never rise above mediocrity, and Intel has shut AMD out of the mid-to-high-end desktop market and almost every segment of the laptop market. AMD has a history of over-promising and under-delivering, and if Intel’s recent CPU launch delays have been disappointing, AMD’s delays are just business as usual.

That said! Given what little we’ve seen about AMD’s upcoming Zen architecture (now called “Ryzen,” pronounced rye-zen), AMD might conceivably challenge Intel at the mid-to-high end of the market for the first time in years. AMD is boasting that a high-end Zen chip with eight cores and sixteen threads can hold its own with the $1,000 Intel chip we highlighted above. If AMD can deliver a full range of dual-, quad-, hexa-, and octa-cores alongside an appealing AM4 platform that provides PCI Express-based storage and integrated 10Gbps USB 3.1 gen 2—and if it can do so while significantly undercutting Intel on price—AMD could definitely regain some credibility with the enthusiast crowd.

The rise of Ryzen is obviously a good thing, even if you don't want to buy into an AMD platform. If Intel feels threatened, it will be quicker to drop prices and introduce new chips to counter AMD’s. Intel just hasn’t had a reason to feel threatened in a while.

But even if AMD never fulfills its promises, Intel still has one big competitor in this space: itself.

If you pay attention to the slides in Intel’s presentations and its message in recent ads, you may have noticed that the company hardly mentions AMD at all. It doesn’t spend much time comparing its newest chips to the previous generation, either. What the company focuses on overwhelmingly is PCs from about five years ago. The PC market is saturated, which means most computers sold are replacing an older one. What Intel has to do is convince you that you need a new computer, and none of the chips from the last two or three years has made a great argument for that.

Processors are complicated, and two to three years normally elapse between the start of development and shipping them to consumers. Let’s assume that, as of 2013 or even 2014, Intel assumed it could still rely on the same tick-tock engine that kept things marching forward. Intel wouldn’t need to overhaul its lineup to keep selling chips because it would be able to improve speed and power consumption enough that there would be clear benefits to upgrading every three years or so, give or take a year.

Now say that somewhere between the official announcement of the Kaby Lake refresh in 2015 and the public death of the tick-tock model in early 2016, Intel saw the writing on the wall and decided it couldn't just rely on improved manufacturing tech to speed up its chips. Assuming a two-to-three year lead time for new architectures, it makes sense that we won’t see six-core mainstream desktop chips from Intel until 2018. Intel just didn’t know that it would need them to keep incentivizing upgrades until pretty late in the game.

Whatever the catalyst ends up being, there’s still life in the desktop CPU. Plenty of people get by with fanless convertible laptops and lightweight Ultrabooks, but people in STEM fields, graphic design, publishing, photo and video production, gaming, app development, and elsewhere can still benefit from high-end processors that are both relatively speedy and relatively affordable. For an unusually long period of time, that sweet spot has been occupied by quad-core CPUs from Intel. But in the next year or two, more competition and realigned priorities ought to get things moving again.