Intel’s Skylake Core i7-6700K reviewed: Modest gains from a full Tick-Tock cycle

Test setup

To test the new Skylake Core i7-6700K processor, we ran an extensive set of benchmarks, and then compared the results against the chip that it ostensibly replaces: the Haswell Core i7-4790K. Rather importantly, the Core i7-6700K has DDR4 memory and the newer Z170 chipset, while the older Core i7-4790K has to make do with the Z97 chipset and DDR3 memory.

Skylake vs. Haswell test systems
	Skylake	Haswell
CPU	Intel Core i7-6700K	Intel Core i7-4790K
Motherboard	Asus Z170 Deluxe	Asus Z97-A
GPU	Intel Graphics HD 530	Intel Graphics HD 4600
Memory	16GB Corsair Vengeance 2666MHz	16GB Crucial Ballistix Elite 1866MHz
Storage	512GB Samsung SM951 M.2 SSD + 256GB Samsung 850 Pro
Cooler	Corsair H100i GTX
PSU	Corsair RM750i
OS	Windows 8.1 64-bit

We’ve used a range of synthetic and real-world benchmarks, including Cinebench, Geekbench, a Handbrake video encode, and the built-in benchmark tools in various games.

The motherboard we used for the Skylake chip—the Asus Z170 Deluxe—will have a UK street price of around £230 (probably ~$300/€300). This is obviously rather dear for a motherboard, with others such as the Asus Z170-K carrying a more reasonable MSRP of £89 (~$120/€120). The Z170 Deluxe justifies it higher pricing with a few extra features, such as USB 3.1, an adapter to add a second M.2 device, built-in 802.11ac 3x3 Wi-Fi, and an SSD secure erase function built into the motherboard’s UEFI BIOS.

The Skylake Core i7-6700K was tested both at its default 4.0GHz clock frequency, and overclocked. The Haswell Core i7-4790K was only tested at its default 4.0GHz clock speed.

We tested a range of DDR4 memory speeds, from 2133MHz up to 3200MHz, along with M.2 performance on the two platforms using a 480GB Samsung SM951.

Skylake CPU performance

Starting with Cinebench 15, Skylake shows a gain of 76 points in the CPU test, which is just under a 10 percent boost. The Cinebench CPU test reflects both the single-threaded performance of a chip and the benefit of its additional cores, so a quad-core processor usually scores about twice that of a dual-core chip, while an eight-core processor would score about twice as highly as the quad-core.

Skylake's power consumption figures at both idle and load are slightly higher than Haswell, perhaps reflecting the higher CPU TDP. For load testing, we ran Prime 95 and noted the power consumption after a a full five minutes. Geekbench 3 results show a measly gain in single-threaded performance of just 140 points, but the multi-threaded test shows a gain of 1300 points; an improvement of around 8 percent.

For a slightly more real-world performance test, we encoded a 15GB MKV video file in Handbrake using the iPad preset and all other settings at their default values. Here the Core i7-6700K shows a great improvement, completing the task 14 minutes faster than the Core i7-4790K, showing a 16 percent gain. The first impressive result.

Skylake iGPU performance (Intel HD 530)

It’s highly unlikely that anyone would buy a top-end Skylake processor and rely on integrated graphics alone for gaming; most people will buy an Nvidia or AMD discrete graphics card instead. But that doesn't mean the new HD 530 iGPU isn't important in the grand scheme of things: this is a key part of the Skylake platform, and will thus almost certainly show up in various laptops over the coming months, such as Apple's late-2015 MacBook Pro refresh.

One day, integrated graphics might be capable of 1080p gaming in high detail on ultra-portable laptops, which also offer great portability and battery life. That could also be bad news for AMD, Nvidia, and vendors of gaming laptops.

But, you may be unsurprised to hear, they can rest easy for now. While the Intel HD 530 iGPU shows strong performance gains in some tests over Haswell’s HD 4600, it’s still nowhere near the performance of even a mid-range discrete GPU.

In the Cinebench OpenGL test, the HD 530 storms ahead of the HD 4600 with a gain of 17 FPS; an improvement of almost 50 percent. The very demanding Fire Strike section of 3DMark showed a gain of 200 points, or 25 percent. Note that this highly detailed scene still rendered at around 2 FPS though, so it’s not exactly indicative of playable performance. OpenCL performance sees solid improvements as well.

Gaming results are mixed: Tomb Raider showed a gain of 5 FPS at 720p, equivalent to a 14 percent improvement, but Bioshock Infinite showed a much better improvement of 17 FPS, or 43 percent. As always, it depends on the game and how much it relies on the CPU or GPU, but clearly the HD Graphics 530 is significantly better than the HD 4600.

Skylake discrete GPU performance

For the sake of completionism, and primarily to sate our curiosity, we also tested the same games using an MSI Twin Frozr 5 GeForce GTX 970 graphics card. The move from Haswell to Skylake made little difference, if any. Frame rates hovered around the same point, whether at 1080p or 1440p resolution, and regardless of the detail setting.

This is a bit disappointing, but it confirms what many PC builders have known for years: for improved gaming performance, a new GPU is the most important component upgrade you can make. A faster CPU will not drastically affect frame rates alone.

Memory performance

With a greatly increased clock speed over DDR3, the addition of faster DDR4 memory is partly responsible for some of Skylake’s improved benchmark results. The Geekbench 3 multi-core memory score indicates how much of an improvement you get from opting for faster memory, while our Handbrake video encoding results slightly improve with faster memory, too.

M.2 storage performance

Both the Z97 and Z170 chipsets support M.2 storage, but you get twice as many PCIe lanes with Z170 (four vs. two), and they're much faster PCIe 3.0 lanes rather than 2.0.

At the moment, the Samsung SM951 is the fastest M.2 drive around—and the only M.2 drive that can get anywhere close to saturating the massive amount of bandwidth available on the Z170. In time, expect to see a lot more M.2 PCI Express SSDs on sale. We were also surprised to see that the 512GB version of the SM951 is only about £30 dearer than the 512GB Samsung 850 Pro: a pretty good deal, if you pick up one of the new Z170 motherboards.

Running the Windows 8.1 setup from a USB stick, we were able to install the operating system onto the SM951 without any problems. On the Haswell (Z97) system we had to change a few EFI settings to boot from an M.2 device. With the Z170 Deluxe, M.2 devices are listed in the EFI without requiring any settings to be changed.

The M.2 performance difference between Z97 and Z170 is enormous. With Z97, we managed read/write speeds of around 780MB/sec; an improvement over SATA SSD speeds, but not by much. On the Z170, the M.2 drive managed monstrous read/write speeds of 1740MB/sec and 1450MB/sec respectively—more than double the performance, using exactly the same drive. 1740MB/sec is about three times faster than state-of-the-art SATA SSDs.

The real-world performance improvement from using a PCIe M.2 SSD isn't quite as dramatic as the synthetic results show, though. Doubling the throughput doesn’t mean a system that’s twice as responsive or boots in half the time.

Skylake overclocking performance

We tried a range of overclocked speeds and voltage settings, keeping an eye on temperatures, before finally settling on 4.8 GHz and 1.4V. This is an increase of 20 percent over the default clock frequency and a good result for any modern CPU. We managed this with both traditional multiplier overclocking and by increasing the value of the base clock (BCLK).

Higher speeds were attainable, but not stable. The Core i7-6700K booted into Windows without a problem at 5GHz, but it crashed as soon as we tried to run a benchmark. Running the chip at 4.9 GHz, it crashed during a run of Prime 95.

This compares quite favourably with Haswell’s potential for overclocking. With Haswell, a boost to around 4.7 GHz could be reached, but going up to 5GHz was really tricky without a really superb cooling setup and tweaking a lot of advanced settings. It’s worth noting that with Skylake the approximate voltage required for the higher overclocks is 1.4V, while on Haswell it’s about 1.3V.

With 4.8GHz, we had settled on a voltage of 1.39V, and every test completed without a problem, except for the Handbrake video encode, which crashed once. Increasing the core voltage to 1.4V solved this, with temperatures of 92°C, which are manageable with a decent all-in-one liquid cooling system. You’ll probably get different results depending on your setup, and experienced overclockers will certainly be able to push the Core i7-6700K further.

At 4.8GHz, Skylake really shows a great improvement over Haswell at its default clock speed: the Cinebench CPU test showed a gain of another 120 points, taking its overclocked result to almost 200 points higher than Haswell, or 23.7 percent faster. The Geekbench 3 single-threaded result shot up 600 points higher, to over 5000, a gain of 15 percent. Multi-core results were almost 4000 points, or 25 percent, better.

With Skylake appearing to have better overclocking headroom than Haswell, this could potentially be a very interesting chip for overclockers.