Fastest ever desktop CPU
Why you should upgrade
You’ll never need a graphics card again
8-PAGE REVIEW FULL BENCHMARKS
Intel’s recent rollout of the seventh-generation Kaby Lake CPU for desktops has met with a muted response from leaked reviews that dismiss the new chip as one huge Core i-Yawn. Kaby Lake seems to offer barely any movement forward, and when it’s overclocked, it apparently gets to nuclear-fusion levels of heat output. It is, however, too early to write off Kaby Lake. There’s a lot more to it that you still need to know.
What Kaby Lake brings to the desktop
Kaby Lake launched in August 2016 with dual-core versions for laptops that offered reasonable performance upticks. The highlight is its video engine, which can handle 10-bit content without breaking a sweat. Play a 10-bit colour depth file on a Skylake laptop with integrated graphics, and you drop frames and destroy battery life. The same video on Kaby Lake hums along with far less impact. You can see this demonstrated at (tinyurl.com/z4tr9bL). The updated graphics core with the latest content protection can now stream 4K from services such as Netflix.
On the desktop side, however, power users don’t care about integrated graphics, focusing more on the lacklustre x86 performance. To be fair, Intel set the expectation in August that Kaby Lake was basically Skylake on an improved process that squeezes out more megahertz.
For example, the top-end Core i7-7700K has a base clock speed of 4.2GHz and a Turbo Boost clock speed of 4.5GHz, versus
a Skylake Core i7-6700K’s base clock of 4GHz and Turbo Boost of 4.2GHz.
The cache size, the core count, the memory controller and even the same LGA1151 socket are unchanged from the previous chip.
This is the real launch
Intel fleshes out the Kaby Lake line-up with a total of 42 CPUs: 17 ultra low-power chips for laptops, two quad-core Xeons, seven quad-core laptop CPUs, and 16 desktop CPUs.
Of particular interest in that desktop line-up to do-it-yourselfers are the three unlocked K chips. The first two were expected: a quad-core with 4.2GHz Core i7-7700K with Hyper-Threading and a quad-core 3.8GHz Core i5-7600K without Hyper-Threading. The third is a surprise: the dual-core 4.2GHz Core i3-7350K. The CPU has Hyper-Threading but since it is a Core i3, does not have Turbo Boost enabled.
This isn’t Intel’s first ‘budget’ overclocking chips, though. The company introduced the dual-core Pentium G3258 Anniversary Edition in 2014 and as early as 2010, Intel sold the dual-core Core i5-655K.
The new Z270 is ‘Optane-Ready’
With Kaby Lake for desktops, Intel is introducing new 200-series chipsets to replace the 100-series chipsets that were introduced with Skylake. Like Kaby Lake, it is an incremental update that disappoints a bit.
We expected the 200-series chipsets to feature native support for USB 3.1 10Gb/s or maybe even Thunderbolt 3, but no. Instead, motherboard makers will have to add additional chips for those functions.
From what we can tell, there are three key changes to Z270. The first is an upgrade from the 20 lanes of PCIe Gen 3 in the Z170 to 24 lanes in the new performance Z270 chipset. The move will let motherboard makers integrate highbandwidth connections such as M.2 or U.2 without having to share bandwidth between devices. Intel says it has also improved overclocking capability.
The last upgrade is official ‘Optane-ready’ support. What that means isn’t exactly clear, but we do know Intel’s Optane (a nonvolatile memory that promises much higher performance than SSDs) will go into an M.2 slot on the board, where it can be
used as a traditional storage device or as a way to accelerate system performance, much like what’s done today with Intel’s Smart Response Technology, which uses a solid-state drive (SSD) to cache performance from a traditional hard drive.
That doesn’t mean Optane won’t work in other systems using older chipsets, but Intel is likely to support it only for ‘system acceleration’ on Z270 initially.
If none of these sound like much of an upgrade to your existing Z170 motherboard, the good news is you don’t have to buy a Z270 mother-board. Kaby Lake drops into most LGA1151 Z170 motherboards and works just fine, as long as you’re using an updated BIOS that supports the new CPU.
How we tested
For our performance testing, we focused on how Kaby Lake does against the CPU it replaces: we dropped the Core i7-7700K into the same Asus Z170 Deluxe motherboard that the Skylake chip was first tested with. The CPU was cooled with a Corsair H80i closed liquid cooler and outfitted with 16GB of Corsair DDR4/2133 RAM, a reference GeForce GTX 980 card, and a 256GB HyperX SSD. The operating system was Windows 10 running the TH2 build.
Cinebench R15 multi-core performance
Our first test is Maxon’s Cinebench R15. It’s a benchmark based on Maxon’s professional Cinema4D rendering engine and is a pure CPU test. We recorded scores from many of Intel’s high-end quad-core mainstream chips and from chips with more cores, for context.
Among the quad-cores, the Core i7-7700K was the winner by the expected amount. The Kaby Lake CPU is roughly 4- to 5 percent higher in clock speed and roughly 4- to 5 percent faster in Cinebench. When you look back to the Core i7-2600K though, it’s a huge 42 percent difference in performance. Stock clock performance between the Kaby Lake, Skylake and Devil’s Canyon though, isn’t exactly going to set the world on fire.
Cinebench R15 single-core performance
One error with focusing exclusively on multi-core performance is reality doesn’t match that. The vast majority of applications are lucky to exploit more than a single thread, instead favouring higher clock speeds and more efficient CPU cores. Once we set Cinebench R15 to run in single-threaded
mode, the quad-cores with their higher clock speeds jump to the front of the line, with the Core i7-7700K now leading the pack. For most people, those who don’t do 3D rendering or other heavily multithreaded tasks, a quad-core with higher clock speeds is the right choice.
There isn’t a lot of daylight between the Core i7-7700K and the Core i7-6700K. Note, too, that our 10-core Broadwell-E Core i7-6950X was performed without Turbo Boost Max. Turbo Boost Max lets the CPU greatly increase the clock speed on a single core, bringing performance a lot closer to the quad-cores.
POV Ray performance
Another CPU-heavy test we use is POV Ray. It’s a ray-tracing program that traces its roots back to the Amiga. Our benchmark set is a little smaller but no surprise, the Kaby Lake, with its 4- to 5 percent clock speed advantage, finishes the test about 4- to 5 percent faster, putting the Core i7-7700K just barely behind a six-core Ivy Bridge-E Core i7-4960X.
As with Cinebench R15, we also run the test using a single-threaded workload. With lighter loads the CPUs can run at higher clock speeds and no surprise, the pair of quad-cores take the lead. Again, we saw maybe a 5 percent advantage for the Core i7-7700K over the Core i7-6700K.
Our final rendering test uses Blender 2.77a and Mike Pan’s BMW work file to measure how fast the various CPUs can render a single frame using the free and popular Blender app. The Core i7-7700K again pulls ahead of the Core i7-6700K by a small percentage, well within what we expected for its clock-speed advantage. And yes, that six-core Ivy Bridge-E Core i7-4960X is really starting to look mouldy here. One thing we’d like to point out about Blender is that it doesn’t show the scaling with thread count as much as Cinebench R15. While the 10-core Core i7-6950X is the winner here, it’s not as impressive as we would have expected for a £1,600 CPU.
HandBrake 10.2 performance
Turning to video encoding, we used the popular and free HandBrake 10.2 encode to convert a 30GB MKV file using the Android tablet preset. The Core i7-7700K again comes in slightly ahead of the Core i7-6700K. There’s also a pretty healthy distance between the Kaby Lake chip and the still-excellent Devil’s Canyon chip. The older Ivy Bridge-E Core i7-4960X disappoints yet again, especially considering that it has six cores yet is basically tied with the quadcore Kaby Lake chip.
You should have the idea by now that the 4- to 5 percent clock increase from Skylake to Kaby Lake pretty much yields a 4- to 5 percent increase in performance across the board, so our last CPU-only test is WinRAR. Unlike the other tests, where we ran the exact same app our machines, these results include both 5.21 and 5.31 results (for the Core i7-5775C and Core i7-4790K). The only difference between 5.21 and 5.31 appears to be bug fixes that don’t impact the built-in benchmark. Unlike Cinebench, POV Ray, or Blender, WinRAR is a little more sensitive to memory bandwidth.
Both the Skylake and Kaby Lake CPUs are pretty much dead even here. We also see the Devil’s Canyon chip is more than 15 percent slower than the Skylake and Kaby Lake chips. The surprise, for the quad-core CPUs, is the Core i7-5775C Broadwell CPU. Despite its lower clock speed of 3.3- to 3.7GHz, it’s leading the pack of quad-core chips.
That isn’t some magic of the Broadwell micro-architecture, though. It’s likely due to the large amount of embedded DRAM cache Intel put into the CPU.
Gaming in this day and age is still 90 percent about the GPU, which is why we ran 3DMark Fire Strike. All of the machines used reference GeForce GTX 980 cards and the same driver. As you can see, it’s mostly a tie. The 10-core Broadwell-E gets a small advantage because 3DMark factors the physics performance into the overall score, but this is mostly a tie.
Because this is a CPU review, we also decided to break out the physics performance, which favours core count over clock speed. No surprise, the 10-core comes out on top. If you’re looking at these two charts and trying to decide how they should influence your buying or building decision, we’d say the graphics score is far more important so long as you have a decently powered quad-core chip.
Tomb Raider performance
We also fired up the slightly older Tomb Raid and ran the built-in benchmark at 1080p resolution at the Normal quality setting. We chose Normal rather than Ultra to try to make this more about the CPU than the GPU.
The Core i7-7700K again leads the pack for quad-cores but it’s really no big deal.
The IPCs have it
CPU nerds like to talk about IPC, or instructions per cycle of a CPU. It’s one way to gauge efficiency at a given clock speed. We took the performance of each CPU running the Cinebench R15 test in single- threaded mode with Turbo Boost switched off on all of the CPUs. As we said with the Skylake Core i7-6700K review, it’s a pretty sobering wake-up call to see just how slowly IPC is inching along in modern CPUs.
The good news for modern processors is IPC isn’t the only place you can pick up performance. Clock speed, core count, and ability to hold Turbo Boost speeds longer have all added up to better performance. Here’s that reminder seen in the first chart we ran from Cinebench R15, when each CPU is allowed to run unfettered rather than locked down to a fixed frequency.
Many of the early unsanctioned reviews of Kaby Lake gave it a black mark for generating excessive heat when overclocked.
We have always been reluctant to offer judgment about the overall overclocking performance of an entire CPU series when working with a sample of one. Combine that with new motherboards, new BIOSs and the dizzying amount of mistakes a reviewer can make, and you can see why we think it’s unfair to decide on an entire line based on one CPU and early motherboards.
Still, in an attempt to get a feel for how Kaby Lake will overclock for most, we spoke to two PC OEMs and a motherboard maker who have been trying to overclock the CPU for far longer and with far more samples.
The Kaby Lake results they’ve seen were quite good. Many of their chips hit 5GHz
or got very close. Motherboard maker Asus, in fact, will feature overclocking profiles that should make overclocking a lot simpler.
“Through rigorous testing, Asus engineers have fine-tuned a profile that allows Kaby Lake CPUs to overclock to 5GHz with an 80 percent success rate,” the company said. This is actually a great sign for practical overclockers because 5GHz overclocks haven’t been seen since the days of the Core i5-2500K and Core i7-2600K. Both of which could seemingly run at 4.5GHz on air or 5GHz with liquid cooling.
You can’t say the same about the CPUs that followed Sandy Bridge. Ivy Bridge and Haswell both hit walls at 4.5GHz for most people. Devil’s Canyon was supposed to break the 4.5GHz barrier but all we got were chips that could get closer to 4.5GHz but not surpass it. Broadwell didn’t count (it didn’t ship in great volume), and Skylake also hit that same invisible barrier at 4.5GHz.
With its massaged 14nm process, Kaby Lake finally seems to break that magical barrier. To prove it, we had two system builders send two production PCs that could break the 4.5GHz barrier. Both did. The first system, for example, was able to withstand almost four hours of continuous HandBrake encodes with all cores locked at 5GHz without issue. The second machine could hit 5GHz in a small-form-factor box.
Running at 5GHz, the Kaby Lake will match a six-core Ivy Bridge-E in performance. In single-threaded applications at 5GHz, the results are even more impressive.
Does this mean your chip will hit 5GHz? No. Remember, it’s always been a lottery system with overclocking results, but the word from experienced boutique PC builders and Asus is far more promising than it’s been in a long time.
As much as everyone wants to be a hater, it’s looking very much like Kaby Lake, for those who want to go there, can break 4.5GHz at last.
Umm, how much again?
So we have an official, sanctioned view of just how a desktop Kaby Lake performs. Now, what everyone wants to know is how much. There is, again, more disappointment.
The initial prices the press was given for Kaby Lake CPUs would have made the Core i7-7700K, at around £300, the cheapest Core i7 ‘K’ CPU the company has ever produced. It was low enough that we mapped out the price of the chip in a chart and was prepared to write that the CPU price war had already begun with AMD over its upcoming Ryzen.
Alas, it was all wrong. Intel updated its price sheets, increasing the cost to £339. That’s the same price as the Haswell, Devil’s Canyon, and Skylake Core i7 processors launched at.
In defence of Intel, every new chip in the price list went up by around £30 to £60. Even laptop CPUs, where Intel essentially has no competition from AMD today, increased. So maybe this was truly just an across-the- Excel formula error and not a reason that’s spelled Ryzen. Clearly, though, the price war with AMD isn’t kicking off with Kaby Lake.
Running at 5GHz, the Kaby Lake will match a six-core Ivy Bridge-E in performance. In single-threaded applications at 5GHz, the results are even more impressive
So let’s sum it up. In laptops, the performance bump is very decent, with perhaps 20 percent or more going from just Broadwell to Kaby Lake.
Desktops aren’t constrained by thermals and battery life the way laptops are, so the performance difference between the generations is far less. The one really big difference between previous chips is the greatly improved video engine. To performance-oriented desktop users, though, integrated graphics – outside of NUC-style mini-PCs – is unimportant.
The price, though equal to Skylake, is a little disappointing for those who expected it to be cheaper, but it’s not like you’re
paying more for less performance. Instead, you’re paying the same price to get a little better performance. Kaby Lake is better and faster, but despite the greater overclocking potential, you can see why, for most DIYers, it’s a little bit of a yawn. Still, some builders should consider it, and we break down the decision tree CPU by CPU right here.
If we had a Core i7-6700K system: We wouldn’t upgrade to Kaby Lake, and we don’t think Intel expects you to unless you want to help prop up its bottom line. There’s just absolutely nothing compelling that would warrant it on a discrete graphics system right now. If Intel’s Optane emerges as a game changer, then you’d consider a move.
If we were going to build a new Core i76700K system: We wouldn’t. Instead, we’d build one using the new Core 7-7700K. Even if you don’t intend to overclock it at first, the stock clock is already higher, and prices will be the same once initial demand settles down. The simple answer is Kaby Lake is better, so there’s no reason to buy Skylake.
If we had a Core i7-4770K or Core i7-4790K system: We probably wouldn’t upgrade. The Core i7-4770K is still a powerful and useful CPU. The only reason would be the need for more M.2 or U.2 storage options, or if you want to be ready for Optane.
If we had a Core i7-4960X or Core i7-3960X system: Even a once mighty six-core CPU can now be matched by Intel’s new Core i7-7700K in some workloads. However, if you were the kind of person who bought a six-core Sandy Bridge-E or Ivy Bridge-E, you care about core count for a reason. It makes far more sense to buy into Intel’s Broadwell-E platform to run a six-core or eight-core CPU. Or just wait to see if AMD’s Ryzen can give you the core counts and performance you need.
If we had a Core i7-3770K or Core i72600K: Look, there’s nothing wrong with the classic Core i7-3770K or Core i7-2600K in actual CPU performance. The problem is your chipset. The Z77 chipset only has two SATA 6Gb/s ports, and good luck trying to run a modern M.2 NVMe drive in them. These platforms are about as creaky as a Austin Maestro with 275,000 miles on the milometer and a leaky gearbox. It’s basically time to upgrade, and Kaby Lake would be fine for both.
If we had a Core i5: You can get by with a quad-core without Hyper-Threading, but anecdotal reports say the days of a quadcore only CPU are drawing to a close. And if you have to upgrade your Sandy Bridge or Ivy Bridge Core i5 chip (or even Haswell or Skylake), it probably makes sense to upgrade all the way to a new Kaby Lake CPU.
If we had an AMD FX-9590: Well yes, an upgrade to Kaby Lake for your AMD ‘eightcore’ would be very nice. But let’s face it, there’s a reason you’re rolling one of AMD’s top CPUs – you’re an AMD fan. Just wait to see if AMD’s Ryzen materializes and offers the price-to-performance ratio people are hoping for, so you can continue to fly the white, black and green flag. If Ryzen does falter out of the gate (unlikely), then, yes, a shiny Kaby Lake might be in your future.
Clock speeds, core count, and process keeps performance moving along
With all five CPUs locked at 2.5GHz, you can see that the efficiency of each has only slowly increased
In an actual game, most of your dividends still come from the graphics chip once you have a decently powered quad-core chip
When it comes to physics there is an advantage, but most gaming is still primarily a graphics task
The CPU still plays very little role in most graphics chores
Asus said its new Z270 motherboards should be able to overclock Intel’s new with an 80 percent success rate
The rear of a Kaby lake CPU
Blender 2.77a is a free render application popular with indie filmmakers
Cinebench R15 puts the new Core i7-7700K at the top of the heap for mainstream quad-core chips
POV Ray mostly matches our results from Cinebench R15. The Kaby Lake chip eases out in front of the Skylake chip, but it’s not a game changer
Just like with Cinebench, when you run on a single core, the higher frequency chips win
The Z170 chipset was a big step forward for mainstream motherboard chipsets, and the Z270 inches it a little bit farther forward
The Z270 is the first ‘Optane-ready’ chipset from Intel. We just don’t know what exactly that means yet
From left to right: an 8-core Core i7-5960X, the new Core i7-7700K, the ancient Core i7-2600K, and a Core i7-4790K
Kaby Lake gets the seventh-generation name because of the improved graphics core and video engine, while the x86 cores are essentially unchanged
Kaby Lake is the first K chip in several generations that seems able to hit 5GHz
A Kaby Lake running at 5GHz in single-threaded tasks will be tough to beat
The initial price of the Core i7-7700K we were given would have been the cheapest Core i7 ‘K’ CPU in history. Unfortunately, that turned out to be wrong
Kaby Lake is a drop-in replacement for Skylake. I’m just not sure anyone should or would do that