First test of Intel’s Skylake platform
Intel’s tick-tock strategy has become slightly less predictable recently. The idea is that, in alternating years, Intel first shrinks the lithography of its chips, boosting efficiency, then brings out a new architecture the following year with a focus on performance. The cycle historically repeated every two years. However, the move to 22nm transistors with Ivy Bridge was over three years ago, which is where desktop processors have remained since. The shrink to 14nm with Broadwell was so heavily delayed that we ended up with a Haswell refresh, called Devil’s Canyon, last year, and now Broadwell has already been superseded by Skylake, Intel’s new 14nm CPU architecture.
As with previous new architectures, Skylake introduces a new CPU socket. This time it’s called LGA1151, adding one extra pin to LGA1150, and breaking backwards compatibility with existing motherboards. As with the move to Intel’s Haswell architecture, the new socket is 100 per cent compatible with existing LGA1150 coolers though.
With Haswell, Intel shifted its focus away from performance gains, looking instead to offer better power efficiency and integrated GPU performance, which was particularly important for mobile chips, as it helped Intel compete with ARM. Skylake is similar in this respect.
The CPUs are certainly a bit faster than their predecessors, but the biggest changes are the new DirectX 12 Intel HD 530 GPU, and the new chipset, (Z170), which adds plenty of new features, including 20 PCI-E 3 lanes (in addition to the 16 lanes in the CPU). The fully integrated voltage regulator (FIVR) that moved on-die with Haswell has been put back on the motherboard as a separate component for Skylake once again, and there’s also now support for DDR4 memory and a move from Direct Media Interface (DMI) 2 to DMI 3, while M.2 slots now get access to four PCI-E 3 lanes, rather than PCI-E 2 lanes, improving SSD performance.
The CPUs
Intel’s consumer desktop processor launches have become fairly predictable, with the specifications of the new Skylake processors almost exactly matching the previous generation. The multiplier-unlocked Core i7-6700K and Core i5-6600K are the two high-end mainstays of the new Skylake lineup. Both are quad-core parts, with the Core i7-6700K offering Hyper-Threading, with each physical core splitting its resources across a second virtual core, enabling it to process two threads at once, which isn’t offered by the Core i5-6600K.
The Core i7-6700K has a base clock of 4GHz, which exactly matches the Core i7-4790K Haswell processor, although it has a Turbo Boost frequency of 4.2GHz, which is slightly lower than the 4.4GHz of the Core i7-4790K. Meanwhile, the Core i5-6600K has a 3.5GHz base clock and a 3.9GHz Turbo Boost frequency.
Both chips have a TDP of 91W, significantly higher than Broadwell’s 65W (although these CPUs have much lower clock speeds), and slightly more than the 88W of the Devil’s Canyon chips. As before, the Core i7 chip has 8MB of L3 cache while the Core i5 CPU has 6MB.
As before, the K-series chips are the only ones with unlocked multipliers, and they also allow adjustment of the base clock (B-Clock) for overclocking.
If you haven’t been paying attention to every recent change in Intel’s CPU architectures, Haswell introduced an alternative way to overclock processors, which was pretty much restricted to multiplier-only overclocking on Ivy Bridge chips. Haswell introduced tuning ratios that overclock the base clock frequency. With Haswell, this was achieved using predefined ratios: 1.0, 1.25 and 1.67, giving you three base clock values from which to choose: 100MHz, 125MHz or 166MHz.
With Skylake, however, the base clock can be increased in much smaller increments to allow for much finer tuning of the clock speed. Although it depends on your motherboard, we found that both methods work well for CPU overclocking. You can leave the multiplier alone and increase the base clock, then manually adjust the memory and cache speeds. That last part is quite important; by increasing the base clock, you affect more than just the CPU. It’s a system that improves the ability to fine-tune your system. With a 40x multiplier, a base clock increase of 0.5 MHz adds just 20MHz to the overall CPU speed, but add an extra 1x to the multiplier and you increase the CPU frequency by 100MHz.
Performance
Since the default clock speed of the Core i7-6700K is identical to that of the Core i7-4790K, with the same 4GHz stock frequency, the benchmark results indicate the exact gains in terms of clock-for-clock performance, and it’s modest to say the least. The overall result in Custom PC RealBench 2015 test was 17 per cent higher than the result from our original Core i7-4790K Devil’s Canyon reference PC, with a score of 133,939.
A part of that increase is down to the fact that our original test GPU bit the dust, though, so our Skylake test rig used a GeForce GTX 970 rather than the GeForce GTX 780 in the Devil’s Canyon machine. This GPU change increased the OpenCL score by 50 per cent, which slightly affects the overall result.
Looking at the score breakdown gives you a better understanding of the bigger picture. In our image editing test, which is mainly single-threaded and responds well to clock speed increases, the Skylake score is just 5 per cent higher, with a result of 63,527, compared to 60,850 in the Core i7-4790K. However, the heavy multi-tasking result of 162,326 is 14,000 points higher than the 148,078 of the Core i7-4790K, representing a gain of just under 10 per cent.
However, there was a more significant gain in our heavily multi-threaded video encoding test, with the Core i7 Skylake chip’s score of 297,115 being around 25 per cent higher than the reference score of 231,750. Skylake is clearly very good at multi-threaded work, and it’s also helped along by DDR4 memory.
Meanwhile, the Core i5-6600K results were still solid, but noticeably behind the 6700K, with an overall score of 105,124. The image editing result of 53,785 was 10 per cent lower than the reference score from the Core i7-4790k, because of the 6600K’s lower clock speed. Meanwhile, the heavy multi-tasking and heavily multi-threaded video encoding results were a fair way behind both the 4790K and the 6700K, as this chip doesn’t have Hyper-Threading, and because of the lower stock clock speed.
Overclocking
In the run-up to Skylake’s launch, leaked benchmarks and CPU-Z screenshots have showed early chip samples to be capable of some extremely high overclocks, with total figures above 5GHz reported using just air cooling. However, there’s usually no mention of system stability or CPU temperatures at these speeds, and a system only needs to boot for a few seconds in order to get a CPU-Z screenshot. With overclocking, you need to know when to stop, and the final frequency on which we settle is the one where the system is stable in Prime95, with temperatures that are within safe limits.
Skylake launched just a few days before we went to press, so we haven’t had time to really probe what’s possible by adjusting every last setting, but we have a pretty good idea. With our test rig, using a Corsair H80i GT all-inone liquid cooler, we achieved a maximum stable frequency of 4.8GHz from the Core i7-6700K, with a 48x multiplier, a 100MHz base clock and a 1.38V vcore. We also achieved the same speed by leaving the multiplier at 40x and raising the base clock to 120MHz.
We also tried a multiplier of 50x with a 100MHz base clock, but Windows wouldn’t boot. However, we managed to boot into Windows with a 125MHz base clock and a 40x multiplier, resulting in an impressive clock speed of 5GHz. At this frequency, though, our benchmarking software crashed as soon as it was loaded. With a 122.5MHz base clock and a final frequency of 4.9GHz, the system was mostly stable, and we were able to run benchmarks, but Prime95 test crashed after a while, and we experienced a few system lockups too.
By default, the Asus Z170 Deluxe motherboard we used for testing automatically sets voltages to what it deems the required levels to reach your chosen overclock, and in the case of our 4.8GHz speed, it set the voltage to 1.48V, which is a little overzealous. While the system was totally stable, the CPU temperature hit 100°C at load, which is simply too high for comfort. Thankfully, the voltage was an overestimation, and with the same settings at 1.38V, the CPU temperature dropped to 89°C under load.
Sadly, our Core i5-6600K sample wasn’t as overclockable as its Core i7 counterpart, despite our best efforts in the time available. We achieved a maximum stable frequency of 4.6GHz (46x 100MHz), using a vcore of 1.35V with this chip.
These overclocks pushed up the benchmark scores, but the most impressive overclocked result came from Cinebench with the Core i7-6700K, where its score of 1,053 is now snapping at the heels of results from 6-core chips, again showing the strength of Skylake’s multi-threaded performance.
Power consumption
The other factor is power consumption, and as this test was conducted just before we went to press, we were unable to test the Skylake chips with the same power supply we usually use for CPU testing. As such, the total system power consumption figures can’t be directly compared to those from other CPUs we’ve tested, but they can still give you an indication.
With either chip installed, our test system drew less than 70W from the mains when idle, increasing to around 150W at load. It looks as though Skylake chips draw slightly less power than Devil’s Canyon CPUs at idle, and around the same amount at full load, which is a solid result when you consider the improved performance. However, when overclocked, the 6700K system drew a much more significant 267W from the mains.
Conclusion
Skylake is a classic story of evolution rather than revolution. It’s an improvement, but based on the modest gains to processor performance at default clock speeds, it isn’t worth upgrading if you already have a Z97 motherboard and a top-end Core i7-4790K. If you’re using an older platform, though, such as a Sandy Bridge PC, there’s a fair amount to be gained from Skylake, especially if you overclock the Core i7 variant and upgrade to an M.2 SSD.
The overclocked results from the Core i7-6700K are great, particularly in multi-threaded software, and the Z170 chipset promises significant performance from PCI-E M.2 SSDs too.
However, given that Skylake represents not one, but two processor generations after Devil’s Canyon, we’re disappointed that its performance gains are only a small step forward. Given the change in focus for Intel, not to mention the current lack of competition in this segment, we have to wonder if we’ll ever see the giant leaps of the past in the mainstream desktop CPU market again. As it stands, though, the Core i7-6700K and Core i5-6600K may not be huge leaps forwards from their predecessors, but they’re still the best chips in their league, and are now our mainstream desktop CPUs of choice. ORESTIS BASTOUNIS
VERDICT
Not a huge leap forwards over Haswell, but the Core i7-6700K offers decent multi-threaded performance, particularly when it’s overclocked, and both chips are still the best in their league.
SPECIFICATIONS:
SPECIFICATIONS:
Frequency Core i5-6600K: 3.5GHz; Core i7-6700K: 4GHz
Core Skylake
Manufacturing process 14nm
Number of cores Core i5-6600K: 4x physical; Core i7-6700K: 4x physical, 4x logical
Cache L1: 4x 32KB; L2 4x 256KB; L3: Core i5-6600K 6MB, Core i7-6700K 8MB
Memory controller Dualchannel DDR4 up to 2133MHz
Packaging LGA1151
Features SSE, SSE2, SSE3, SSSE3, SSE4, SSE 4.2, EM64T, F16C, Turbo Boost 2, Quick Sync Video