FYI.

This story is over 5 years old.

Intel’s New Skylake Chip Is Good, And That’s the Problem

The next big leaps in performance aren't happening as fast as they once did.
Intel's Kirk Skaugen demonstrates 'Skylake' at IDF14. Image: Intel

Intel's new Skylake CPU is out, and the early reviews are mostly positive, albeit reserved. But while Skylake is great, the problem is it's just not that much greater than Broadwell, the chip architecture released late in 2014 that it replaces. And based on Intel's plans for the next couple of releases, that trend might continue.

In terms of Intel's alternating "tick-tock" release structure—where every tick is a shrinking of the chip and every tock is an improvement on the chip's architecture—Skylake is the (barely audible) tock. The future just doesn't seem to be getting here as fast as it once was.

Advertisement

When Gordon Moore, one of Intel's founders, made the observation that would eventually become known as Moore's law, he predicted the density of transistors on a CPU would double every year. Then it was every 18 months, then every two years, and now, finally, it's every two and a half-years. We've told you about the slowing down of Moore's Law before, and the release of Skylake to modest reviews is more proof that this is the new normal.

Of course chip size and transistor density are not the only thing that make a CPU better, but while Skylake is no slouch, it's not blowing its predecessors out of the water either.

Skylake comes with support for DDR4, the latest version of dynamic random access memory (DRAM), that uses less power while upping data transfer rate and keeping latency about the same. While DDR4 should also make your memory less susceptible to hacks like the (admittedly very cool) Rowhammer exploit you may have heard about, in desktop PCs, tests have suggested users will only see up to a whiplash-inducing 5 percent increase in performance speed.

Its integrated graphics processor is also much better, so the casual gamer may just be able to play games at near-high definition 720p resolution with "medium" settings. But for gamers who plan on actually, you know, playing modern games at 1080p resolutions, they'll still need dedicated graphics cards. Meet the new chip, (mostly) the same as the old chip.

Upgrade fetishists and gamers, the two groups most likely to laugh at the $249 to $350 price range, are getting less bang for their buck than in previous upgrades. In fact, as Ars Technica points out in their review, Intel compared Skylake's speed not to Broadwell—its 2014 release and the tick to Skylake's tock—but to Haswell, a processor from 2013, and another tock. Their next chip, Kaby Lake in 2016, will be yet another tock and the third 14 nanometer chip the company will release. A smaller and denser chip won't be out until the 2017 release of Canonlake.

Intel isn't floundering by any means, but it does seem to be managing expectations for how fast their tech will improve. The modest gains made by Skylake over Broadwell are reminiscent of the small gains Haswell made over its own predecessor, Sandy Bridge. Enthusiasts will always upgrade when they have the opportunity, but it's obvious that the average user can keep their current computer running for longer before it becomes necessary to upgrade.

Something has to give. Users will either have to manage their expectations as we near an almost certain performance plateau, or chip manufacturers will have to blow us away with some new advancement. The laws of electrodynamics being what they are, there's only so much you can increase the density of components on a chip. Moore's Law has been 10 years away from dying for a while now, but that's not the point. Skylake isn't blowing anyone away, and if the next 'tock' uses the same processes, it won't either. We expect technology to increase exponentially, not linearly, and a future where you literally can't buy a "faster" computer next year might be a scary one for some.

Correction: An earlier version of this story referred to clock speed, which is a misnomer, for the transfer rates of the chips. The story also originally included a link to the wrong page of a multi-page review in AnandTech.