Intel's Skylake is officially debuting today, just months after the company launched new high-end desktop Broadwell chips. It's been more than two years since Intel launched a new desktop architecture, and over a year since the company's Haswell refresh, the Core i7-4790K, debuted as a high-end enthusiast chip. The new Core i7-6700K and Core i5-6600K are meant to replace the old CPUs, with new integrated graphics, new chipset support, and an all-new core architecture.
Complete details on that core architecture will have to wait until IDF August 18th, minus what we can glean from early chip tests. Intel isn't sharing much in-depth detail on the chips today. For now, we'll cover some initial results and discuss the platform and its positioning within Intel's lineup.The road to Skylake
The two new Skylake chips debuting today are the vanguard of a full suite of sixth-generation processors coming later this year. Intel hasn't yet described how it will launch the comprehensive hardware refresh, but given how recently Broadwell debuted in some markets, we would assume that Intel will push Skylake out on a similar cadence. That way Broadwell has enough time in market to move inventory and the company doesn't end up stuck with chips it can't sell.
Intel's entire strategy around Skylake has been a little strange. When the company announced that Broadwell would launch up to a year late in some market segments, we assumed this would have an impact on Skylake's launch schedule as well. It only made sense that Intel would hold its cadence timer steady, rather than rushing out a new launch on the heels of the old one. Instead, Intel did exactly that, opting to replace the older Haswell desktop parts rather than do a top-to-bottom refresh based on Broadwell.
At the time, this may have made sense, but it introduced an odd hiccup into Intel's roadmap. The company recently announced that its 10nm ramp would be delayed by roughly a year, with a new interim product, Kaby Lake, brought forward to serve in the interim. The only thing Intel has done with the 10nm ramp is formalize the delay that characterized Broadwell rather than informing investors that it would happen mere months in advance, but the firm could've avoided the need for Kaby Lake if it had simply kept Skylake in the oven for a bit longer.
Either way, we're here now -- with the first new Intel architecture to debut in over two years. Skylake is the follow-up to Broadwell, which makes it a "tock" in Intel's nomenclature. The new chip is also built on 14nm, but unlike Haswell, it uses a new socket, LGA1151. Along with a new core and socket comes a new motherboard chipset, the Z170.
As initial leaks suggested, the new Z170 chipset makes a number of modest improvements to Intel's previous high-end Z97. GPU PCIe 3.0 connectivity is the same, but the total number of USB 3.0 ports has increased to 10, up from 6. Total USB 3.0 ports in total is still 14, though this would seem to address the needs of all but the most crazed USB 3.0 users in general. The major chipset change is the use of DMI 3.0, up from 2.0. This means available point-to-point bandwidth has leapt from 20Gbit/s (PCIe 2.0, essentially) up to 8GT/s (or 8GB/s). That's a huge bandwidth jump for Intel's point-to-point interconnect, and it's how the chipset can feed the 20 lanes of PCIe 3.0 connectivity that the Z170 provides.
Skylake features
There are several new features to discuss regarding the combined chipset + CPU platform. First, Intel is offering full base clock overclocking again for the first time in years. Previously, overclocking via bus speed has been extremely limited on Sandy Bridge platforms or above. While Intel added support for multiples of the FSB, it typically only offered them in set ratios. Now, gamers and enthusiasts will be able to hand-tune processors to a much finer degree than was previously available.
The other theoretical advantage is the support of high-end DDR4, up to and including DDR4-4133. The reason we say "theoretically" in this case, however, is because the overwhelming majority of consumer workloads are latency limited, not bandwidth limited -- and DDR4 latency is sufficiently high compared with DDR3 that the gains are going to be modest. The highest-end DDR4 on the market, DDR4-3400, is timed at 16-18-18-36. High-quality DDR3-2133, in contrast, is timed at 8-10-10-27. DDR4-3400 is clocked 1.6x higher than DDR3-2133, but its listed latency figures are also 1.6x higher. (Actual latency calculations are more complicated than this, so treat these comparisons strictly as a ballpark estimate.)
Relative RAM latency
This is why next-generation RAM standards take so much time to offer concrete overall performance gains compared to previous-gen counterparts. Strictly in terms of latency, the latest modern DRAM standards struggle to move much past DDR-400 with 2-2-2-5 latency. That doesn't mean DDR4 is bad, of course, but don't look to a new memory standard to offer much in the way of additional performance.
The Core i7-6700K and Core i5-6600K
Intel is formally announcing two cores today -- the Core i7-6700K and the Core i5-6600K. Both chips support either DDR4 or DDR3L, with peak speeds up to DDR4-2133 and DDR3L-1600 officially. We've already had our board up to 2667MHz, so clearly there's more headroom on the hardware than the official standard. Each chip is designed to fit within the Core i7 vs. Core i5 paradigm that we discussed a few weeks ago; the Core i7-6700K has four cores and eight threads for $350, while the Core i5-6600K is a four-core, four-thread chip with a 6MB L3. That said, neither chip is a stand-out leader in terms of raw clock speed -- Intel's Core i7-4790K runs at 4GHz with a boost clock of 4.4GHz, while the Core i5-4690K has the same clock speeds as the Core i5-6600K.
This is probably why Intel kept its predictions of performance gains relatively restrained, with a stated "up to 10%" improvement for a one-year-old PC. Intel is claiming roughly 10% a year, meaning that Skylake should be 10% faster than last year's Devil's Canyon, 20% faster than the Core i7-4770K, and 30% faster than the older Core i7-3770K, aka Ivy Bridge. I'm still working on finalizing our review (I've been mostly focused on Windows 10), but here's a bit of a taste of what Skylake offers in terms of performance.
I've included two test results below, but only one of them has a full suite of comparison figures. The x265 test is new, and the only system I had immediately at hand to test against was my own Core i7-3960X. Still, we see some intriguing performance figures from just these benchmarks:
I want to talk about the older H.264 benchmark first, since we've got a healthy set of performance data to choose from here. Note that the Core i7-6700K, shown in blue, is actually faster than the six-core Core i7-3960X in the first pass and within 2.5% of it in the second. That's surprising because we rarely quad-core chips beating past six-core processors, even when the six-cores are from an earlier generation. The 6700K does have a clock speed advantage over the Core i7-3960X, however, since the latter chip tops out at 3.9GHz turbo, compared to 4.2GHz for the 6700K. That 6% clock speed improvement explains some of the Core i7-6700K's performance improvement against the Core i7-4770K, but by no means all of it. Skylake is faster than Haswell by a substantial margin, at least in x264 encoding.
Intel's newer six and eight-core chips still beat the Core i7-6700K, but the gap is narrower than one might expect.
Finally, in the newer x265 encode test, we've only got two points of reference -- the Core i7-6700K and the Core i7-3960X.
This is a new test that I decided to include out of curiosity, and we haven't checked it thoroughly for threading and other capabilities. Even so, it shows a formidable performance gain for the Core i7-6700K. Skylake wins past Sandy Bridge-E by 30% -- a huge gain, particularly for a quad-core processor.
We're working on a full review of Skylake and looking forward to IDF when Intel will shed more light on the architecture of the CPU. Based on what we've seen thus far, however, this is one solid chip.
