Date: January 3, 2017
Author(s): Jamie Fletcher
Intel’s latest 7th generation CPUs were launched a couple months back with a focus on the mobile markets. Today, Kaby Lake finally comes to the desktop with eager hands awaiting. A new platform is introduced with the Z270 chipset, new overclocking features, Optane Memory, and an unlocked i3 CPU. Read on as we put the flagship Core i7-7700K under review!
Back in late August, early September last year, Intel introduced its latest generation Core processors, Kaby Lake. At the time, the focus was on notebook and mobile designs, featuring 4-15W TDP devices. Low power, compact, high performance chips that are readily available in new mobile designs.
You can read up on some of the basic design overviews of Kaby Lake in our previous article, but today marks the release of the latest desktop releases, starting with the headline chip, the Core i7-7700K.
Kaby Lake for desktops brings with it a new chipset to work with, and that entails new motherboard designs and features. For the most part, Kaby Lake is not too dissimilar from Skylake, as they both share the same socket (LGA1151), and as is usually the case with such launches, Kaby Lake is backwards compatible with the previous chipset (Z170, B150 etc), at least with a firmware update. To take full advantage of the updated features though, you’ll need the new chipset, Z270, which we’ll cover in this review too.
The main new features of the platform include an update to Speedstep, native HEVC 10-bit plus VP9 encode and decode, a new SSD trick with something called Optane Memory, and an overhaul to the integrated graphics. There are quite a few Windows 10 specific features too, along with some architecture changes under the hood regarding I/O with PCIe lanes, Thunderbolt 3.0 and USB 3.0.
Finally, there are several changes to the way overclocking is handled, including tweaks to how BCLK works and a decoupling of AVX instructions. All of these tweaks we’ll get into in a little bit. We should point out that a number of these features are for Kaby Lake in general, while others are specific to certain products within the generation.
The new Speedstep or Speed Shift (depending on where you look) is an extension of Turbo Boost 2.0 that changes the way turbo frequencies and idle states are handled. Previously, CPU speed scaling was handled by the OS or software, dropping frequencies when idle to save power. Speed Shift passes control from software to hardware.
This shift to hardware control means faster switch speed… much faster. Under software control, stepping was enabled after 15-30ms, the new hardware control drops this down to 1ms. There aren’t any details about the number of steps and what frequency ranges these include, but the net result is a significant improvement to response times when ramping up for heavy workloads, and dropping back down to an idle state to save power.
While these improvements are not going to be appreciated as much on desktop processors, especially if you enable high performance mode in the OS (which prevents sleep/idle states and frequency scaling), it does mean that normal office PCs will see improved responsiveness.
Hardware accelerated video has seen a massive boom in the consumption and generation of media. In the good old days of watching videos on PCs, most of the decoding was done through software via the CPU. For lower resolution content, this was fine, but when 1080p came along, and now 4K, software decoding became impractical, especially for low-end CPUs.
These days, pretty much all CPUs and GPUs have some kind of video acceleration enabled, even on mobile. Thing is, video is changing as more and more people are not only watching 4K streams from the likes of Netflix, but also uploading video in real-time with ‘Let’s Play’ gaming. Not only were new codecs required for more efficient streaming, but HDR video meant that higher bitrates are required.
HEVC or H.265 was introduced a couple of years ago under Intel’s leadership to help mitigate the problems of streaming 4K video over Internet connections. While H.264 has served us well, it starts to become bloated when dealing with 4K content, putting a massive strain on often sluggish and arbitrarily capped Internet connections. H.265 allowed for the same image quality at a much lower bitrate, anywhere from 30-50% less, but at the cost of more processing power.
Video has continued to evolve though, and a new feature is on the block with HDR video, or High Dynamic Range. This takes advantage of a wider color gamut for a significant increase in contrast. Typically though, this requires 10-bit color channels in the video, rather than 8-bit. The new codec support on chip means it can decode these richer color videos. Why is that important? Netflix streaming of UHD content will require this, along with Windows 10 (and a compatible TV). The VP9 codec, which is an alternative to HEVC, also gets hardware integration, but no 10-bit color support. This is mainly for Google’s own YouTube streaming platform.
These gains are not just for decoding video, but encoding too. As streaming from your own computer becomes more common, rendering the game and encoding video at the same time, can put a serious strain on a system. External HDMI capture cards are common not just for consoles, but PCs too. One less component in the chain means lower latency and less to go wrong.
Nope, that’s not a typo, Intel is changing the naming scheme of its graphics cores to simplify some of the technologies. Originally, there was Iris and Iris Pro graphics, which are a separate, but more powerful line of Intel’s HD series integrated graphics. However, some of the waters became murky when EDRAM was introduced.
EDRAM is dedicated on-package RAM that significantly improved rendering speed and some computation workloads of the CPU/IGP. The problem was that there were certain Iris Pro enabled processors with and without this extra RAM. Starting with Kaby Lake, all IGPs with EDRAM will go under the naming scheme of Iris Plus.
Currently, the only supported chips with Iris Plus are the U-Series processors, which are the 15-28 Watt chips. S-Series chips, which are the performance and high-end chips such as the i7-7700K, will only have the HD series graphics, rather than Iris Plus. Whether we are likely to see another Extreme Edition CPU with EDRAM again is anyone’s guess.
In general though, Iris Plus is significantly faster than the HD graphics. By Intel’s own metric, Iris Plus is up to 65% faster in 3D rendering and gaming, and 40% faster using QuickSync video rendering, when compared against the HD 620. But that won’t have much bearing on the desktop parts we’ll be reviewing later on with the i7-7700K.
We’ve heard rumors and hints that a new class of CPU will be released with Kaby Lake, and this has been confirmed; Intel’s very first K-Series Core i3 CPU.
If you are not familiar with K-Series, this is Intel’s naming scheme for overclockable CPUs, so yes, it means for the first time Intel is allowing budget, dual core CPUs to be overclocked with an unlocked multiplier with the Core i3-7350K.
The new Core i3-7350K sits in a peculiar place in the new product line-up, as it’s still an i3 processor and subject to the same limitations, but with a few tweaks. It’s still a dual core chip with SMT (4 threads), but it has a larger TDP of 60 Watts instead of 51 Watts, so there is more thermal headroom. Intel released a dual core Pentium series with an unlocked multiplier a few years ago, but it lacked hyperthreading.
Core i5 and i7 chips have two different frequencies, a base clock and a turbo clock. The turbo is engaged under heavy workloads where thermal and power limits allow. This turbo is different depending on the number of cores that are in use, so a single core task will have a higher turbo than a multi-core task. Since the Core i3 processors only have two cores, they do not have this turbo.
From an overclocking perspective, this lack of turbo simplifies things, although it does mean you can’t set a higher clock speed for single-core processes. Since there are fewer cores in general, there is less heat, so in theory, you should be able to achieve much higher clock speeds compared to something like a Core i5-7600K. However, this chip is something we’ll put to the test in a later article when we get one in to review. We expect to see it launch some time in February or March
If an overclockable dual-core chip got you excited, Intel is pushing its overclocks even more with changes to BCLK and introduction of an AVX Instruction offset ratio, the latter of which we’ll get into in a little bit.
There will be a new menu option for overclockers called B Clock Aware Voltage/Freq Curve. Most users when overclocking, typically only adjust the multiplier, which is a fixed value multiplication of the base clock of 100MHz. So a multiplier of 45 equals a 4.5GHz overclock. For the more adventurous, it’s possible to overclock the base clock for fine tuning. So a base clock of 105, multiplied by 45, is just over 4.7GHz, but achieving that can be tricky.
The new B Clock Aware feature automatically adjust the voltage applied to the CPU to offset the increased base clock. In effect, this increases the stability of the overclock when the base clock is increased. It’s unlikely that there will be any major gains to this, simply because the base clock is used by more than just the CPU, as it’s tied to the PCIe lanes and other I/O. This tweak is mainly for managing the voltage of the CPU when increasing the base clock – which you could perform yourself manually if you wanted.
The last item on the overclock tweaks is a new AVX instruction offset ratio. This is less of a tweak and more of a fix for a problem that has only really come to light in the last year or so. The AVX instruction set is a specialized architecture extension that works with large integers and vector multiplication. For the most part, you won’t come across many applications that use it, but one sector that does, is video encoding.
While H.264 and H.265 continue to get hardware acceleration encoding through the GPU and dedicated encoder units, not all video formats can take advantage of those processing engines. Digital cameras often have their own media formats, and software encoders provide the full list of options available for the encoders, rather than a subset as with hardware-based encoders. Software encoders can provide better quality and compression than hardware accelerated ones.
These software encoders make use of the AVX instruction set to achieve much better performance than by using SSE instructions. The problem though, is that AVX uses more power than normal to complete. This led to a peculiar problem when it comes to overclocking. Under certain workloads, overclocking the CPU and using the AVX instruction set often resulted in less performance than if the chip was left at stock frequencies – not good.
The new AVX offset ratio works as a throttle for the system frequency when AVX instructions are detected. This means that when the CPU is overclocked and it detects an AVX instruction, it will not engage the full overclock, but instead use a lower clock or disable the overclock altogether. This helps keep the CPU within the power envelope as well – which was the cause of the issue in the first place.
If there was one feature that left a few scratching their heads, it would be Optane Memory. Currently, there isn’t a huge amount of information available about the specifics of Optane, all we can go by are some very general ideas. We will of course update this as soon as we are given the specifics.
First of all, the name is somewhat misleading by calling it ‘memory’ when it relates to storage, but it’s a feature that makes use of Intel and Micron’s 3D Xpoint (crosspoint) memory. This is a type of non-volatile storage (it doesn’t forget when powered down) that’s significantly faster than the standard NAND flash-based storage found in solid state devices (thumb drives, SSDs, e.MMC, etc). While it can supposedly get near the speed of DRAM, so far there have been no real-world performance metrics. 3D Xpoint is aptly named for its 3D towering structure of cross-point connected cells, leading to large columns and rows of individually addressable spaces. Specific details on this is where things get fuzzy, although it shares a number of similarities with another technology called memristors.
In terms of the specific implementation of 3D Xpoint in form of Optane Memory in a system, the way we can best summarize it is that it’s a high-speed cache for hard drives (and SSDs). An Optane Memory module plugs into a dedicated M.2 slot on the motherboard and sits above the storage stack, and below RAM. The system gets a huge storage performance boost from the Optane module since it’s almost the same speed as the RAM, without having to configure an additional storage device from the OS. This will likely be different from a pagefile, and probably not visible to the OS.
Again, no specifics as of yet, especially with regard to total storage capacity. Whether the Optane module and system storage are treated as a single storage drive with a combined total, like with a RAID or JBOD array, or if it appears as a large cache that sits above the storage and saves to the hard drive in the background, is also unknown.
The Optane module will be significantly faster than a hard drive, and likely faster still than a high-end SSD, even one of the NVMe PCIe M.2 SSDs currently on the market. How much faster is again, another unknown.
Any 200-series chipset with a Core i3 or above will be able to take advantage of Optane Memory, if not now but in the future, should an M.2 port be made available.
While the above concentrated on certain key specifics, there are a few architecture changes too, mainly to do with I/O and connectivity. For the most part, things are roughly the same as Z170. The same number of USB ports for both USB 2.0 and 3.0 (14 and 10 respectively). There is support for up to 6 integrated SATA 3.0 ports, which is also the same.
Since we are dealing with the same socket type, there is very limited connectivity with which to add more features. However, there is one increase that, on the surface doesn’t means a lot, but is still important nonetheless; that’s an extra 4x PCIe 3.0 lanes to route through the board, bringing Z170’s count of 20, to 24 on Z270.
With some playing around of the lanes, it’s possible to enable an extra 1×16 slot on a motherboard, useful for high-speed PCIe-based SSDs, and to a lesser extent, graphics. However, those 4 extra lanes have a more immediate benefit of allowing an extra x4 enabled M.2 port on the motherboard. This ties in directly with the Optane Memory feature detailed above.
Some motherboards will already have two M.2 slots, a third could be made available… if there is enough physical space without resorting to a riser card. Audio on premium boards have to do this already. More likely is that single M.2 boards under Z170 can be updated to include a second port for Optane on Z270.
Z-Series chipsets are still required for overclocking as well. You can use your existing Z170 board with a Kaby Lake CPU too, but you’ll be missing out on most of the extra features, including the extended overclocking tweaks.
Before we get into the testing, we’ll give you a quick rundown as to the system specifications. We decided to use the same motherboard for testing both the i7-6700K and i7-7700K.
The mothered is the soon to be released ASUS STRIX Z270E (yes, a STRIX Motherboard, finally). Due to time constraints and updates to a number of our test software, we’ve only got a single CPU to test against.
At some point, we’d like to add one of the Extreme Edition CPUs to the testing suite as well. All tests were performed at least twice to verify results, with a third time if necessary when there is a large amount of variance in the results.
Below is a table listing our current test bed.
|Intel CPU Test Systems|
|Processors||Intel Core i7-6700K 4.0-4.2GHz|
Intel Core i7-7700K 4.2-4.5GHz
|Motherboards||ASUS STRIX Z270E Gaming|
|Memory||4x8GB G.Skill Trident Z DDR4 @2133MHz|
|Graphics||6700K – HD Graphics 530|
7700K – HD Graphics 630
NVIDIA GeForce GTX 1060 6GB
|Storage||Crucial MX300 500GB SSD|
|Power Supply||Corsair AX1200|
|Chassis||Corsair Obsidian 800D Full-Tower|
|Cooling||Corsair H100i V2 Extreme Water Cooler|
|Et cetera||Windows 10 Professional 64-bit|
We’re only testing against the 6700K due to time constraints. Later on we’ll take a better look at other CPUs if possible, when we get back from CES. Performance metrics are going to be somewhat limited over the generation, but most of the reason for upgrades comes from features.
As always, we like to start all benchmarks off with an overall view of system performance, and for that we use PCMark. In general we prefer to use SPECwpc, but this time around we ran into issues and crashes. Once wpc has been updated, we’ll update the results, likely when we run our overclock tests.
|PCMark 8 Suite Scores|
|Intel Core i7-7700K||5699||8386||6065|
|Intel Core i7-6700K||5567||8182||6007|
|Higher results are better.|
A boost to the CPU clock speed and we get a boost to the different workloads – not terribly exciting. With a lack of any major architecture improvements and using the same manufacture process, this is what we expect. As we go through the results though, this is something to keep in mind – Kaby Lake isn’t a major overhaul, but a refinement of existing products.
Adobe products have integrated OpenGL support and upped its multi-threaded processing a lot over the years. We use Lightroom to convert 500 RAW images to JPG and time the result. It should be noted that Adobe CC has regularly updated its packages, so these results won’t scale with previous versions.
|Intel Core i7-7700K||414 s|
|Intel Core i7-6700K||438 s|
|Results in seconds; lower is better.|
This follows on from what we were expecting, a linear improvement over the generations. Compared to the previous benchmarks from our older Skylake review, Lightroom has taken a large hit to performance this generation of Adobe Creative Cloud, up to 50% slower. Not much we can do about that, but the results show scaling between Skylake and Kaby Lake all the same. Such is the joy of always updated, cloud-based software.
We have kept the same workload for Premiere Pro from our Skylake testing, which includes 4K video, as well as leveraging Premiere’s new processing techniques that take advantage of GPUs. Since the update to CC 2017, there has been a major increase in rendering performance, and it’s not just Adobe PP showing improvements.
|Adobe Premiere Pro CC|
|4K RED Encode||PPBM9|
|Intel Core i7-7700K||97 s||85 s|
|Intel Core i7-6700K||100 s||91 s|
|Results in seconds; lower is better.|
We use Handbrake a fair bit and these new render settings leverage both the CPU in the x264 and x265 tests, and the IGP for QSV (Quick Sync Video). It still needs to be updated to take advantage of the full profile options available on the GPU, so these results will change over time. While the speed will likely remain the same, later results will typically net better compression or image quality.
|H.264 QSV||H.265 QSV||x264||x265|
|Intel Core i7-7700K||89||91||172||207|
|Intel Core i7-6700K||96||97||186||223|
|Results in seconds; lower is better.|
Intel wanted particular focus on its HEVC/H.265 codec performance and recommended MAGIX Movie Edit Pro 2017 with the extra HEVC codec pack. This takes full advantage of the hardware capabilities and options of the IGP.
|MAGIX Movie Edit Pro 2017|
|Intel Core i7-7700K||89 s|
|Intel Core i7-6700K||96 s|
|x265 HD Benchmark|
|Intel Core i7-7700K||21.15 FPS|
|Intel Core i7-6700K||19.80 FPS|
Any improvements here will be from the IGP and QuickSync engine, rather than the CPU and it’s clock gain. With video rendering out the way, we move on to 3D rendering. For the most part, this is an industry that is still very heavily entrenched in CPU and software-based rendering, so no IGP or GPU acceleration just yet. While rendering engines such as Iray and ProRender are beginning to show promise, workflows are still very much geared to the more expansive options provided by software rendering over the CPU.
|Autodesk 3ds Max 2017|
|Intel Core i7-7700K||744|
|Intel Core i7-6700K||792|
|Results in seconds; lower is better.|
With our Naomi render in 3ds Max, there are general improvements overall, but nothing special. This is largely down to the increase in base clock and turbo, as there has been very little architecture improvements overall that affect CPU performance. Kaby Lake is definitely showing itself as a feature upgrade, rather than a performance upgrade.
|Intel Core i7-7700K||152||1002|
|Intel Core i7-6700K||147||939|
|Higher results are better.|
With Cinebench, it’s much the same. The increase in clock speed shows typical gains to the overall performance. Even with the OpenGL test which uses the GPU, doesn’t see a huge gain either.
|Intel Core i7-7700K||474||2145|
|Intel Core i7-6700K||442||1999|
|Higher results are better.|
|FLAC to MP3|
|Intel Core i7-7700K||676 s|
|Intel Core i7-6700K||711 s|
|500 FLAC to 320Kbps MP3. Results in seconds; lower is better.|
Finishing up the CPU specific performance metrics, we’re back to the expected gains over the generations.
A CPU’s weighting on game related performance is often negligible for modern titles. For the most part, the CPU’s single-threaded performance is the biggest factor. Modern titles are finally beginning to take advantage of more threads and advanced instruction sets, as seen with the likes of Battlefield 1, but in general, clock speed is the deciding factor.
|3D Mark Fire Strike|
|3DMark results in points; higher is better.|
While Intel focuses on 3D Mark’s lower end tests like Sky Dive, realistically speaking, that provides little relevance to modern games (the benchmark is getting on a bit now). If we were testing the IGP for something like a Laptop, then we might include it, but we’ve got an overclockable premium CPU here with dedicated graphics – no IGP here.
As such, the results are mostly predictable. Graphical performance differences are negligible, but physics tests come out ahead with the 7700K – which is expected when there’s an extra 300MHz on the turbo, and 200MHz on the base clock.
|Unigine Heaven 4.0|
|Min FPS||Avg FPS|
|Intel Core i7-7700K||35.3||88.9|
|Intel Core i7-6700K||8.9||88.3|
|Settings: Ultra-tex, 4xAA, Normal-tess|
Results in Frames Per Second; higher is better.
The same is seen again with Unigine Heaven. Don’t pay too much attention to the min FPS of the 6700K as that is just a hiccup that can occur with scoring, as it measures the lowest FPS, rather than a more typical average of the 1% lowest FPS.
With the system specific tests completed, we’ll delve into the sub-system performance with SANDRA. These tests are long and arduous with huge amounts of granularity to them, providing a better understanding of the types of workload the chips are best suited to.
|SiSoftware SANDRA 2016 SP3|
|Arithmetic||123.76 GOPS||132.63 GOPS|
|Multi-Media||420.39 MPix/s||453.81 MPix/s|
|Cryptography High Security||8.72 GB/s||9 GB/s|
|Financial Analysis||24.0 kOPT/s||25.63 kOPT/s|
|Scientific Analysis||24.2 GFLOPS||25 GFLOPS|
|Memory Bandwidth||26.00 GB/s||26.33 GB/s|
|Cache & Memory Latency*||24.1 ns||23.6 ns|
|Cache Bandwidth||178.24 GB/s||186.27 GB/s|
|Higher is better|
*Lower is better
Without much in the way of performance tweaks in the architecture, there isn’t a huge amount of difference between the 6700K and 7700K – which is again, expected. It’s that extra clock speed showing gains overall.
|SPECapc 3ds Max 2015|
|Large Model Composite||5.46||5.67|
|Higher is better|
|Higher is better|
The same again with SPEC tests, which are a mix of both CPU and GPU based workloads. The small differences in favor of the i7-6700K are just a reflection of the natural variance in test results. Viewperf is much more focused on the GPU though, hence the results being almost identical across the board.
|Intel Core i7-7700K||32C||72C|
|Intel Core i7-6700K||31C||74C|
|Results in degrees centigrade; Lower is better|
So where are the surprises? Well, temperature results fall in with what we expect. Since the 7700K uses the same 14nm manufacture process as the 6700K, and has the same TDP, we get pretty much the same thermals. The slightly higher idle is the result of the higher base clock, but the lower load is likely the result of the improved Speed Shift – which we can see in the power results below.
|Intel Core i7-7700K||47W||134W|
|Intel Core i7-6700K||50W||144W|
|Results in Watts; Lower is better.|
Power is recorded at the wall and measures the total system draw. The new Speed Shift and overall optimization to the manufacture process, and we get less total power being pulled by the system. Since the system can clock up and down faster depending on the workload, expect to see some power savings.
Kaby Lake is a difficult CPU to call out any final thoughts. It’s Intel’s stop-gap to 10nm due to the incredible complexity that comes from ever-shrinking manufacture processes. The general result is unfortunately a rather dull performance gain that is entirely down to the increase in clock speed.
Don’t get us wrong, this is still Intel’s most advanced CPU currently available; but for the desktop, it’s a hard sell. Kaby Lake is highly focused on the mobile sector, with very big power efficiency gains that are very hard to test on a desktop. The faster speed it can switch between clock frequencies is not something you’ll easily notice unless you have a battery.
Same again can go for the focus on video encoding and decoding. High-end desktop CPUs have plenty of processing power to decode 4K video in software, and most desktop owners will have dedicated graphics, which themselves have video acceleration too. It’s low-power devices found in notebooks that are really going to benefit from the video acceleration, especially with Netflix Streaming using H.265.
Optane Memory can’t be tested just yet, so we can’t provide any feedback on some of the more specific features of the Z270 platform just yet. Unfortunately due to time constraints, the CPU being delivered over Christmas, and CES just about to start, we haven’t been able to really push the 7700K with a battery of overclock tests.
As soon as we get back from CES, we will be putting the new overclocking tweaks to the test and see if we really can push these chips much higher than Skylake, as well as check out the performance of the AVX tweak. Once the i3 K-series SKU is launched as well, then we’ll definitely be taking a look at that as well.
We’re holding off on a final verdict until we can fully test the overclocking capabilities of Kaby Lake, and put Optane to the test, then we’ll have a better understanding of the desktop market. The tray price per thousand is just over $300 for the 7700K, but expect to see retail price of around $380+. We’ll update when we start seeing them for sale shortly after CES.
Copyright © 2005-2017 Techgage Networks Inc. - All Rights Reserved.