Intel Core i9 is the next generation processor. Processors Review on i9 processors

Intel Core i9-7900X Review | Introduction

Intel's new Skylake-X processors include the Core i5, i7 and i9 family models, but they are all compatible with the LGA 2066 interface and the X299 chipset. The chips are designed specifically for high-performance desktop (HEDT) users looking for a processor with 4-18 physical cores. Meanwhile, existing Skylake-S processors for conventional desktop computers are installed in motherboards with an LGA 1151 socket.

The company claims that a number of Skylake-X architecture enhancements provide 15% performance gains over Broadwell-E in single-threaded workloads, and 10% in multi-threaded workloads.

Intel began to dominate the high-end desktop processor segment about a decade ago, and the processor market has been skewed since then. In the absence of serious competition, Intel did not see the need to cut prices or introduce significant innovation. But AMD recently returned to the game with Ryzen processors offering many cores, SMT, and an unlocked multiplier, all for less money.

Intel now wants to defend the leading position claimed by the recently announced AMD Threadripper processor, which offers 16 cores, 32 threads and 64 Gen 3 PCIe lanes. Of course, the improvements to Skylake-X are not pure feedback, as the technology has been in development for many years. However, Intel announced the new hardware unusually fast (tying your shoelaces on the run). In addition, the company has adjusted prices, and we have not seen such offers for a long time. Enthusiasts, enjoy.

Specifications

Intel Core i9-7900X specifications
Price in USA, $	1000
Price in Russia, rub.	n / a
Connector	LGA 2066
Cores / Threads	10/20
Thermal package	140 watts
Base clock frequency	3.3 GHz
Turbo Boost frequency	3.3 GHz
L3 cache	13.75 MB
DDR4 RAM support	2666
Memory controller	four-channel
Unlocked multiplier	yes
PCI Express Lines	x44

The 12-core (or more) Skylake-X models are still unavailable and the Core i7-7440X for our tests was canceled for no reason. As a result, we are left with a 10-core Core i9-7900X for review. Let's take a look at the entire line of high-performance Intel processors.

Kaby lake-x

We're not used to seeing current-generation architecture in a range of high-performance PC processors. As a rule, top-end models are one or two generations behind mainstream chips. The arrival of a couple of Kaby Lake-based chips for the LGA 2066 socket is changing things. Fortunately for those who love the familiar, the rest of the PCH X299 comes from Skylake, although that may soon change. Earlier this year, Intel announced its new "Data Center First" strategy, in which the latest manufacturing processes are implemented in Xeon processors and only then in desktop products. Considering that the line of high-performance chips consists of repurposed dies for the data center, these chips could become cutting edge.

Intel is unprecedentedly expanding its high-performance CPU family from four to nine models, including two models based on Kaby Lake-X. These two chips support dual channel DDR4 memory, while Skylake-X offers four channels of RAM. This means that with a Kaby Lake-X CPU installed, you can only use half of the motherboard DIMM slots. Fewer PCIe lanes also result in limited I / O capabilities. Intel disables the onboard HD Graphics 630, allowing the unused portion of the die to absorb excess heat, which supposedly improves overclocking. Apart from slightly boosted base clock speeds and a TDP of 112W, the Core i5-7640X and i7-7740X processors are otherwise similar to their Skylake-S counterparts, including prices.

In our opinion, the installation of "inexpensive" processors in expensive motherboards resembles the situation with the Core i3-7350K, which are not popular due to a similar imbalance in price. According to Intel, motherboard manufacturers can build low-cost X299 platforms specifically for Kaby Lake-X, but we don't know of any such products yet.

	Core i5-7640X	Core i7-7740X	Core i7-7800X	Core i7-7820X	Core i9-7900X
	250	350	390	600	1000
Family	Kaby lake-x	Kaby lake-x	Skylake-X	Skylake-X	Skylake-X
Technical process	14nm +	14nm +	14nm	14nm	14nm
Kernels / threads	4/4	4/8	6/12	8/16	10/20
Base frequency (GHz)	4	4,3	3,5	3,6	3,3
	4,2	4,5	4	4,3	4,3
	n / a	n / a	n / a	4,5	4,5
L3 cache size (MB)	6	8	8,25	11	13,75
PCIe 3.0 Lines	16	16	28	28	44
Memory support	2 can. DDR4-2666	2 can. DDR4-2666	4 can. DDR4-2400	4 can. DDR4-2466	4 can. DDR4-2666
TDP, W	112	112	140	140	140
Processor socket	2066	2066	2066	2066	2066
Chipset	X299	X299	X299	X299	X299
Unlocked multiplier	Yes	Yes	Yes	Yes	Yes
	$242	$339	$389	$599	$999

	Core i9-7920X	Core i9-7940X	Core i9-7960X	Core i9-7980XE
	1200	1400	1700	2000
Family	Skylake-X	Skylake-X	Skylake-X	Skylake-X
Technical process	?	?	?	?
Kernels / threads	12/24	14/28	16/32	18/36
Base frequency (GHz)	?	?	?	?
Intel TurboBoost 2.0 frequency (GHz)	?	?	?	?
Intel TurboBoost 3.0 Frequency (GHz)	?	?	?	?
L3 cache size (MB)	?	?	?	?
PCIe 3.0 Lines	?	?	?	?
Memory support	?	?	?	?
TDP, W	?	?	?	165
Processor socket	2066	2066	2066	2066
Chipset	X299	X299	X299	X299
Unlocked multiplier	Yes	Yes	Yes	Yes
Wholesale price (1000 units in the USA)	$1,199	$1,399	$1,699	$1,999

Skylake-X

All Skylake-X processors except the Core i7-7800X support DDR4-2666 memory. Broadwell-E only officially supports DDR4-2400. The manufacturer deliberately disables ECC to dissuade potential Xeon processor users from buying enthusiast platforms.

Intel did not provide detailed specifications for high-end processors, but we expect memory support to be different. We also expect clock speeds to drop as core counts increase.

Active nuclei	1	2	3	4	5-10
Intel Core i9-7900X Turbo Boost Clock (GHz)	4,3	4,3	4,1	4,1	4,0

Like the previous generation processors, the Core i9-7900X is equipped with Turbo Boost 2.0 technology, but at higher clock speeds. With ten cores active, 4 GHz can be expected. Intel also equips six Skylake-X models with Turbo Boost Max 3.0 technology. The company has improved this technology and now accelerates the two fastest cores for low-threaded tasks. In Broadwell-E, Turbo Boost Max 3.0 accelerates only one core. Both target cores exceed 4.5 GHz. Naturally, the throughput of instructions per clock should increase, which was sometimes lower in the highest-performance Intel chips than in the quad-core versions. Today, Turbo Boost Max 3.0 requires a separate driver on some motherboards. But Intel plans to implement support for this technology in Windows 10. Partial support for AVX-512 is also implemented, so the future 18-core Intel flagship will be the first desktop CPU with a computing performance of more than 1 TFLOPS.

Skylake-X differs markedly from Skylake-S with a completely redesigned cache structure. The Core i9-7900X has more L2 cache and less L3 cache, which should improve performance in most applications. Mesh 2D architecture also appeared. By analogy with AMD Infinity Fabric, this element of the microarchitecture does not give an extremely positive effect (more details later). Enthusiasts were very upset to learn that the 10-core Core i7-6950X will sell for $ 1,700, but the manufacturer estimates the 10-core Core i9-7900X at $ 1,000. For this thousand, you get 44 PCIe 3.0 lanes, and the Core i7-7820X only gives 28 lanes. Multi-GPU configurations are not so popular today, but storage systems are gradually moving to PCIe and additional lanes may come in handy for connecting SSDs. Intel also introduced a new PCIe Virtual RAID on CPU (VROC) feature that allows up to 20 SSDs to be combined into a single boot volume. Notably, you can build a RAID array on any available PCIe slot, although previous versions of RSTe RAID required a chipset connection. Bypassing the chipset removes the DMI bottleneck. Unfortunately, it is not given free of charge. You will need to purchase an upgrade dongle that plugs into the motherboard to unlock the VROC feature. Server users are familiar with this practice, but it won't be popular with enthusiasts. We don't even know how much the key will cost.

Intel is bringing back overclocking for the DMI and PCIe bus, which should please users. The new PLL voltage setting is designed to increase the memory overclocking potential based on the factor, and the new AVX-512 factor offset combines the standard AVX offset for temperature control when processing demanding AVX tasks. While testing the Core i9-7900X, we encountered some anomalies in processor performance. The launch of the new processor was clearly in a hurry, and while the motherboard firmware updates (from several vendors) solved some of the oddities, some of them still survive. It looks like the Intel Skylake-X chips will take some time to optimize, as will the AMD Ryzen processors. Let's take a look at the factors that affect Skylake-X's performance.

Intel Core i9-7900X Review | Altering the matrix

In the course of testing, we encountered rather strange trends and inconsistencies in terms of performance. Considering Skylake-X's clock speed advantage, updated cache and 2D mesh topology, we didn't expect Broadwell-E to have a chance to win. However, in some cases, the previous generation flagship outperformed the Core i9-7900X. We asked Intel to comment on the situation, and the response was the following: "... we noticed that there are several applications in which the Broadwell-E processor is comparable or superior to the Skylake-X processor. This is due to the differences in" mesh "Skylake-X and in" ring "Broadwell-E architectures.

Each new architecture requires engineers to make technological compromises in order to improve the overall performance of the platform. Skylake-X's mesh architecture is no different from others in this regard.

While these tradeoffs affect several applications, overall, the new Skylake-X processors offer excellent IPC speeds and dramatically improve performance across a wide variety of applications. "

Our American colleagues have already reviewed the mesh architecture of Intel for Skylake-X and Xeon processors (English). See this article for details. Of course, there is not enough information to complete the picture - many data have not yet been disclosed. However, we are seeing a significant change in efficient processor design, so it should come as no surprise that the mesh topology does not provide additional performance across the board.

Prerequisites

Interconnects are channels for moving data between key components within a processor, including cores, tier caches, PCIe controllers, and memory. They affect latency and power consumption, which in turn affects performance and heat dissipation.

Intel's ring bus debuted in 2007 in Nehalem processors, and AMD's HyperTransport technology was introduced in 2001. Both technologies have evolved, but increased cores, more cache, and high I / O bandwidth have put a strain on the interconnects. To achieve greater performance gains, there are several ways to increase interconnect speed, although this often requires higher data rates and therefore higher voltage.

A good example of a solution to this problem is Intel's bidirectional ring bus, marked in red above on a Broadwell die with few cores. Data travels to components in a circular path, and as the number of cores increases, latency increases. The second picture shows a 24-core Broadwell crystal. Aligning building blocks on a single bus is impractical in this case due to high latencies, so Intel split the large die into two separate ring buses. This solution complicates scheduling, and buffered switches that communicate between buses add five cycles of latency, limiting scalability.

Against this backdrop, AMD introduced Infinity Fabric with the Zen microarchitecture, which is currently implemented as two quad-core processor units communicating over a 256-bit bidirectional bus, which also serves Northbridge and PCIe data. They also share a common memory controller. Route through the Infinity Fabric bus to another quad-core CCX and its cache results in increased latency. We've detailed the design and measured its latency in our review. We also found that higher memory clocks can improve the latency of the Infinity Fabric bus, which is likely one of the key reasons Ryzen's performance increases along with higher memory transfer rates.

AMD claims that software and platform optimizations can fix some of the performance oddities we noticed in our testing. Based on what we saw not today, we can conclude that this is true. AMD's efforts and a string of BIOS, chipset, and software updates have resulted in significant performance improvements over our first Ryzen 7 review.

AMD's work continues. And now Intel is facing the same problem.

Mesh structure

The 2D mesh architecture debuted in intel Knights Landing Products ... The network consists of horizontal and vertical interconnections between cores, cache, and I / O controllers. There are no buffered switches in the circuit, which have a very negative impact on latency. The ability to "stagger" data through the cores allows for much more complex and supposedly efficient routing. Intel claims the 2D network has a lower voltage and frequency than the ring bus, while still providing higher bandwidth and lower latency.

Similar to the Knights Landing design, Intel has moved the DDR4 controllers to the left and right sides of the 18-core die. They used to be at the bottom of the ring bus crystal. Based on the Skylake-X die shot, it has six memory controllers (second row, bottom right and left), so Intel disabled two controllers by default. The company is likely using a smaller LCC die for the Core i9-7900X, although company representatives are not giving exact information.

Mesh characteristics

CPU	Latency time inside the core	Core to core latency	Average core-to-core latency	Average baud rate
Core i9-7900X	Core i9-7900X	69.3 - 82.3ns	75.56 ns	83.21 GB / s
Core i9-7900X @ 3200 MT / s	16 - 16.1 ns	76.8 - 91.3ns	83.93 ns	87.31 GB / s
Core i7-6950X	13.5 - 15.4 ns	54.5 - 70.3 ns	64.64 ns	65.67 GB / s
Core i7-7700K	14.7 - 14.9 ns	36.8 - 45.1ns	42.63 ns	35.84 GB / s
Core i7-6700K	16 - 16.4 ns	41.7 - 51.4 ns	46.71 ns	32.38 GB / s

Measuring the latency inside the core shows the latency between threads that are on the same physical core, and the metrics from core to core reflect the latency of threads between threads of two physical cores. The Core i9-7900K is most comparable to the 10-core Core i7-6950X, but for comparison we used the quad-core models.

Intel strikes back. The developer's response to AMD's Ryzen Threadripper is two incredible, overflowing processor cores: the 18-core Core i9-7980X and the 16-core Core i9-7960X.

However, did Goliath-Intel really recover from the recent crushing defeat inflicted by David-AMD? Are the unflattering rumors about clock speeds and CPU overheating disproved?

Gordon Ma Ung, the executive editor of PCWorld magazine, figured out this one of the recognized experts in "hardcore" testing. He also tested the performance of the new Intel Core i9 chips under real-world conditions to see if the price was worth paying for them.

Since there is a lot to talk about Core i9, prices, bells and whistles and answers to the most obvious questions this time we will leave aside. In our review, we will go through some of the more subtle aspects of performance that are not so obvious, and then dive into benchmark comparisons.

Intel Core i9: what is hidden under the hood

The Core i9 is the first new "Core i" processor released by Intel in the past 10 years. The company kept the secret so zealously that it even deliberately mislabeled the first batch of chips, signing them "Core i7" in order to confuse the traces of leak seekers. Nevertheless, our 16- and 18-core prototypes are correctly signed.

CPU-Z thinks Core i9 is Core i7

Like most of Intel's global developments, the Core i9 family introduces not just a new processor, but a completely new platform, which means a completely new X299 chipset, as well as a new LGA2066 socket that is incompatible with previous processors.

The new platform also provides something that no one has ever done before, unifying the two processor families. Previously, if you were looking for a Kaby Lake chip, it required an LGA1151 socket motherboard. If, say, you wanted to buy a 6-core Skylake, that is, Intel Core i7-6800K, you had to buy a motherboard with a V3 base and an LGA2011 platform in the kit.

With the X299 boards and the LGA2066 socket, you can make your choice after purchasing the motherboard, as this platform supports all new CPUs from the 4-core Core i5 Kaby Lake to the 18-core Core i9 Extreme Edition of the Skylake line. To be clear, the Kaby Lake series, also called Kaby Lake-X series processors, include the new Core i5-7640X and i7-7740X chips. The rest of the Core i7 and Core i9 chips belong to the Skylake family, collectively referred to as Skylake-X.

The Core X series consists of processors consisting of Skylake-X cores and Kaby Lake-X cores. The 18-core monster from this line was released in October

We expect this union with some confusion and concern. It looks like X299 motherboards are going to be quite expensive. I’m curious who would want to buy a $ 350 motherboard to install a $ 250 processor on it.

Intel's motives for continuing the Kaby Lake-X line may actually be a nod to overclockers. Unlike older Kaby Lake processors for the LGA1151 socket, the new Kaby Lake-X chips do not have integrated graphics on board. In fact, they are physically devoid of integrated GPUs. This will allow the two new Kaby Lake-X processors to overclock potentially significantly higher than the LGA1151 versions. At a recent Computex show in Taipei, Intel announced that the highest overclocking record was set on the Kaby Lake processor and X299 motherboards.

In an ideal world, we would all have 18-core processors, but the truth is that there are indeed people out there who buy relatively cheap processors for high-end motherboards. Kaby Lake-X is just for them.

PCI Express buses: distribution by coupons

Still, the location of Kaby Lake-X and Skylake-X on the same socket is somewhat disappointing. The most compelling argument is the allocation of PCI Express lanes. For example, with the Core i9-7900X chip, you get quad channel RAM support and 44 PCI Express Gen 3 lanes directly from the processor. If you decide to attach a Core i7-7740K to this connector, the motherboard will drop memory support to two channels. And perhaps even worse, the number of PCI Express lanes will be reduced to 16, since this is the maximum supported by the Kaby Lake cores. Whence it follows that some slots on the motherboard will give up in performance or completely stop working.

While Kaby Lake's 16-lane limit is processor-dependent, Intel's Skylake-X PCI Express lanes are deliberately reduced. While the 10-core version also gets 44 lines, only 28 lines are already available for the 6- and 8-core variants of the Skylake-X. As far as we understand, there are no technical reasons for this - there is a clear “segmentation of the market”, which translated from business language into a common language means “so we can rip more money from you”. Oops.

You may need to buy a special dongle dongle if you want to use the X299 VROC option to activate RAID up to 20 NVMe drives.

Intel VROC

Even more dubious than PCI Express allocation is another option from Intel, VROC, or Virtual RAID on the processor. This is a great Skylake-X feature that allows up to 20 NVMe PCIe RAID drives to be assembled into a single boot segment.

What is the problem? Intel apparently intends to squeeze even more money out of the users of this option. The exact details are still unknown, but vendors at Computex believed that RAID 0 will remain free, RAID 1 will cost $ 99, and RAID 5 and RAID 10 will cost users $ 299. After paying the required amount, the user will receive a special dummy key that will unlock this option.

And even worse, VROC will work exclusively with Intel SSDs and more expensive Skylake-X CPUs. After purchasing Kaby Lake-X, you are dropped from the game. VROC is also only applicable to PCIe RAID, which can be connected directly via PCIe processor lanes. The X299 continues to support RAID 0, 1, 5, 10 options across the chipset, but the chipset RAID will not have any effect on the performance provided by VROC.

AVX 512 in the Skylake-X series promises more performance - but only if the code supports it

How the Core i9 is changing the Skylake series

After overcoming the confusion and disagreement about the platform, you still get a hefty reward. The Skylake-X processor itself is an admirable thing as it is designed in a slightly different way from previous high-end consumer processors.

Previous CPUs, be they "enthusiasts" or "extreme", were basically similar in design. For example, the 4-core Haswell Core i7-4770K does not differ much from the 8-core Haswell-E Core i7-5960X, with the exception of support for 4-channel RAM.

With Skylake-X, Intel is breaking this tradition by introducing extremely significant design changes. The most noticeable is the increase in the Mid-Level Cache (MLC) cache, or L2 cache: Intel brought it to 1 GB per core, raising it four times against the 256 MB of last year's Broadwell-E models and most of Intel processors. The Last-Level Cache (L3), meanwhile, is getting smaller, 1.375 MB per core versus 2.5 MB of the previous Broadwell-E chip, but Intel compensates for this loss with an increased MLC cache, as well as using a non-inclusive cache design. Compared to Broadwell-E's inclusive design, which can continue to store data that is no longer needed, the non-inclusive cache tries to keep track of what is worth saving, so it promises to make better use of available space.

Skylake is very different from the previous Skylake-X line, and the matter here largely depends on the AVX512 cache and the new mesh architecture

Intel is also changing the ring bus architecture that has been in use for several years (including Kaby Lake and Skylake) to a new mesh architecture. Imagine a 4-core processor as four houses connected by a bus line that stops at every house. This all works great as long as there are 12 to 18 houses in the area. It is possible to launch two bus routes, but still it will not be as fast as just moving from one house to the next, which is implemented in the new mesh architecture.

The ring bus architecture of recent processors has been retired in favor of a mesh architecture that promises to provide better speed for a large number of cores

Intel's use of a honeycomb design clearly puts the company in a better position to compete successfully with Threadripper as more and more cores are added to processors. AMD in the Ryzen series is taking advantage of what the company calls Infinity Fabric, which is essentially a super-high-speed mesh network.

And the last feature worth mentioning is the improved Turbo Boost Max 3.0. Intel recognizes the "best" best processor cores at the factory and gives them a little more extra speed. On Broadwell-E processors, only one core is selected. In the Skylake-X series, two cores are already labeled as "best" and can run at a speed of a couple of hundred megahertz faster.

Nuclear War: Episode IV (can you find the error in this picture?)

18-core Core i9 performance

For performance testing, we pulled a 10-core Core i9-7900X from its socket on an Asus Prime X299-Deluxe motherboard and put an 18-core Core i9-7980X in there. Other components in the test suite include a GeForce GTX 1080 Founders Edition graphics card, 32GB DDR4 / 2600 RAM, and HyperX 240GB Savage SATA SSDs. For our Adobe Premiere CC 2017 test, we used a Plextor M8pe PCIe SSD as both the starting point and the destination drive, in all cases except for the Core i5 and Ryzen 5 processors. We had to make an exception for them due to a problem with the motherboard under Ryzen 5, which flatly refused to recognize the Plextor drive. Instead, I had to use the Samsung 960 Pro NVMe SSD. AMD Ryzen Threadripper 1950X remains the same that was originally used by us to write a review of this chip, where it was tested on an Asus ROG Zenith Extreme X399 motherboard, with an Nvidia GeForce GTX 1080 graphics card, Samsung 960 Pro SSD and 32 GB of DDR4 / 3200 RAM ...

Due to the time limit, some benchmarks captured data from the Core i9-7960X processor, the 16-core version of this chip. The processor was used on a pair of identical Falcon Northwest Talon systems assembled specifically for the planned Threadripper vs. Core i9 test. Although these systems are equipped with completely different graphics processors, this does not affect the operation of the system processors in any way, so the data on them can be quite comparable.

Cinebench R15 performance

Our first benchmark is CineBench R15, a free 3D rendering benchmark based on the professional Maxon Cinema4D engine. It is almost entirely tied to the computer's CPU, and is also very sensitive to an increase in the number of cores and processes.

The winner, perhaps, is not a surprise: this brainchild of Intel, the 18-core Core i9-7980X, its smaller brother, the 16-core Core i9-7960X, took second place. AMD's Threadripper 1950X, until recently the undisputed leader among consumer CPUs, had to settle for bronze.

However, there is nothing shameful for the Threadripper 1950X in third place. Yes, AMD fans, yes, we know and remember: its cost is significantly lower. Let us immediately announce it publicly, so that you can calmly read our review to the end without the constant desire to shout: "But it is several times cheaper!" Just repeat this phrase to yourself after you see the results of each test, okay?

Cinebench R15 awards 18-core Core i9 gold medal, 16-core Core i9 silver and AMD's Threadripper 1950X bronze

But multi-streaming activity is far from the salt of the earth. The sad truth is that the vast majority of programs and applications simply don't use all of these cores, so we also ran our chips through the CineBench benchmark to measure single threaded performance. And here we are in for a surprise: the Core i9-7980X processor comes out on top again, outperforming the even more overclocked Core i7-7700K. For the most part, we see three tiers of performance here, with Kaby Lake and Skylake-X chips at the top, followed by Broadwell processors and other Zens.

Just to keep things perspective: we're not currently looking at the huge difference between Skylake-X and Broadwell-E or Zen processors. But the winners in this competition are definitely the Core i9 and the Skylake-X series.

Evaluating single-threaded activity with Cinebench R15 is valuable for predicting how the processor will handle the vast majority of games and applications.

Performance in POV Ray

The Persistence of Vision Raytracer actually traces its history back to the days of the Commodore Amiga, and it continues to be supported by an active developer community. Like Cinebench, it also favors multi-core and high-threading chips. The test results are quite predictable, with the 18-core Core i9-7980X at the top of the list. The 16-core Ryzen Threadripper 1950X performed well enough, but a couple of extra cores are paying real dividends.

Since we would still like to know how the processors will behave under a much lower load, we run the POV Ray test on a single thread. And again, high-speed chipsets of a quadratic architecture come up, but Skylake-X chips are almost catching up with the leader, and Zen with Broadwell-E is practically breathing down the neck. The only really lagging behind here is AMD's already outdated Vishera-based FX processor.

POV Ray 3.7 places the fastest chips with the highest interprocess communication in the first places in the list of results

Performance in Blender

Our next test is the free Blender 3D modeling software. It is a popular application that is used to create effects in many indie indie films. Blender's performance results can vary greatly depending on the task being performed. For example, the performance of some tests carried out on a 4-core Kaby Lake from Intel and Ryzen from AMD is practically independent of the number of cores. For the same task, we ran Mike Pan's popular BMW test file. Again, Intel's two new Core i9 CPUs were the winners, closely followed by the Threadripper 1950X.

Again, all three main processors in our study have excellent performance. And again, the speed indicators in Blender are very dependent on both the chip model and what we actually do with it. In addition, we found that Blender was quite sensitive to the operating system.

Blender's open source renderer also prefers the processors with the most cores. Mike Pan's popular BMW test file was used here

Since these are really fancy chips, we decided to test them with something more complicated, for example, a test file from Gooseberry Production. This is a control frame from the Blender Institute's forthcoming movie Space Laundry. While a BMW task takes only a couple of minutes to complete, Gooseberry loads the electronic brain with the work of processing a frame for a good 20 minutes.

The Gooseberry results on our Falcon Northwest Talon systems look great for the new Core i9s and definitely paint the worst picture for the 16-core Threadripper 1950X.

Gooseberry pushes Intel's new Core i9 processors way ahead of AMD Threadripper 1950X

Performance in WinRAR

We know from our original reviews of the Core i9-7900X and Threadripper 1950X that WinRAR doesn't seem to be particularly inclined towards the mesh architectures of these processors. So it won't be a surprise for us to see the same picture now, although it was quite unexpected to see how much older Broadwell-E chips outperformed them. Alas, Threadripper did not perform well here.

The popular RARLab WinRAR archiver doesn't particularly like the mesh architecture of the Skylake-X series, but it seems that he just hates the Zen architecture of AMD

7-Zip performance

We also used version 9.20 of another archiver, free 7-Zip, to run a built-in multi-thread test on it. The clear winners, breaking away from the rest of the list by a wider-than-expected margin, were the new Core i9 processors.

The free and popular 7 Zip again moves the most multi-core chips to the first positions

Performance in Corona Renderer

Looking at the results from Cinebench, Blender and POV, the difference in performance between the 16-core Threadripper and the new Core i9 is visible, albeit small. In the results of testing with the Corona Renderer, we observe such a gap, which is simply breathtaking. The 16-core Core i9-7960X beats its 16-core Threadripper 1950X by a 25 percent lead. For the 18-core Core i9-7980X, the difference is even greater.

Before someone yells that the benchmarks were deliberately chosen to glorify Intel's microarchitecture, we hasten to state that this particular study was suggested to us by AMD specialists for our original Threadripper review. To be honest, this graph looks very so-so.

Corona Render reveals 16-core Threadripper wins out with 16- and 18-core Core i9 processors

Performance in Handbrake

Not every future user is involved in 3D modeling, but a lot of people edit or convert video files, and this is exactly the area where the multicore processor is most useful. To evaluate the encoding performance of the new Core i9s, we used the popular and free Handbrake encoder to process a 30GB 1080p video file using native Android tablet presets.

We would like to draw your attention to one curious aspect that we encountered when analyzing the results of this study. The more the number of processor cores increases, the more the gap between file processing times is reduced. You can see for yourself how steeply the performance grew while we were moving from 4-core to 10-core chips, but after this milestone the speed gain became extremely insignificant, at least not as much as we would have expected on 18 cores.

Once again, both Core i9s processors are ahead, although this time the Threadripper also shows quite decent speed.

The results of our tests with the Handbrake encoder also confirm that more cores contribute to better performance, but still not as much as a professional 3D renderer would.

Performance in Premiere Creative Cloud

The other half of video processing is, of course, editing. For this specific test, we chose Adobe Premiere Creative Cloud 2017 and real footage from our video department projects, so this testing is as close to real conditions as possible. This footage was captured with a Sony Alpha camera at 4K and exported with a Blu-ray preset at 1080p. We also set the rendering quality to the maximum level, which helps to keep the image level high when changing resolutions.

Although this task is mainly processor-intensive, we have made some efforts to ensure that other components do not interfere with the comparison. Therefore, for all systems except Ryzen 5 and Core i5, we used a Plextor PCIe NVMe SSD as the data source and destination drive. As in the previous testing by the Handbrake program, the file processing speed depending on the number of processor cores does not decrease in direct proportion, although the 18-core Core i9 still remains the champion.

However, if you are buying a powerful processor for editing video files, you should carefully consider how much the overpayment for the number of cores will bring in speed.

Snobs will tell you that processor-based rendering is the most important and most difficult task, so if you do it, you need more cores

And one more thing that we would also like to add. Many would argue that in the age of GPU encoding, system chips don't really matter. To prove or disprove this claim, we reconfigured Adobe Premiere from CPU processing to GeForce GTX 1080 GPU processing with CUDA technology. As you can see, using the GPU immediately yields a huge boost in speed, but the increase in the number of processor cores is clearly paying off as well. And it would be strange to think that a dual-core processor is better at video editing than a 10-core one.

Even if you use the GPU for transcoding, more cores on the system chip significantly reduce the processing time of video files

Rise of Tomb Raider performance

Stop. If you're buying a 16-core or 18-core processor primarily for PC gaming, you're doing it wrong. Much wiser would be to spend that money on a more advanced graphics card. But if you also do 3D modeling in addition to gaming ... and are wondering which processor will give you the best performance ... we suspect you already know the answer: of course, the Core i9.

We say this because we already know how good both the 10-core Core i7-6950X and the 10-core Core i9-7900X are for gaming. The new Core i9 models don't break this once established order.

The first exploration game was Rise of the Tomb Raider, refined to be effective on Ryzen and Threadripper platforms. We ran the game at 1920x1080 and medium settings in DirectX 11 mode.

The 18-core Core i9-7980X is again at the top of the leaderboard, but for the most part its results are not too far off the 10-core Core i9-7900X. Threadripper performs quite well in Game Mode, but it still fails to overtake the Core i9 even in this case.

Intel's Skylake-X series continues to deliver the best performance in most PC games, but Threadripper 1950X is still in the game.

Performance in Tom Clancy's Rainbow Six Siege

In fact, we tested several games on our processors, but for the most part, the 18-core Core i9-7980X was at the top of the list, at times very close to the first place. We saw a similar trend in Tom Clancy's Rainbow Six Siege, launched at medium quality at 1920x1080. We chose these settings in order to eliminate the impact of limitations on the capabilities of the video card on performance testing.

Core i9 scores top in Rainbow Six

Performance in 3D Mark Time Spy 1.0

Our last gaming test is the 3D Mark’s Time Spy 1.0 test. Only the chip fraction is taken into account, since nothing else interests us at the moment. Once again, the power of the Core i9-7980X remains undeniable.

3D Mark's TimeSpy again places the 18-core Core i9-7980X at the top of the list, although it is obvious that the numbers here are not at all directly related to the number of cores.

Energy consumption and speed

What else interests us about the Core i9-7900X is its power consumption, as well as how much more power it uses compared to AMD. This is usually not the easiest question to answer due to the different test equipment, but this time, as we noted earlier, Falcon Northwest sent us two nearly identical Talon system units packed with state-of-the-art components for research. Both are equipped with 128GB DDR4 / 2400 RAM, Samsung 960 Pro SSDs and Titan Xp SLI graphics cards, and they have the same power units, coolers and cases. The only difference between these system units is motherboards and processors.

This kit allows us to measure the energy consumed by the processor on different tasks right on the socket. Since most of the test tasks do not actually load all the cores, we decided to take measurements while increasing the load from one to 32 threads. The results confirmed what everyone already knew: Core i9 consumes more power.

Using a pair of nearly identical 16-core systems, AMD's Threadripper 1950X has been confirmed to be more power efficient than its competitor, Intel's 16-core Core i9-7960X.

These energy measurements are not entirely accurate, but close enough to give us an interesting idea. Curiously, the Threadripper 1950X seems to freeze at 20 threads, while Core i9 data continues to climb.

Threadripper certainly has a power consumption advantage, but that's not the most important factor. When multi-threaded performance is extremely important to you, a couple of extra kilowatts spent are unlikely to matter to you.

This is quite similar to the gaming performance of Threadripper. Yes, of course, the advantage of the Core i9 is undeniable, but honestly, hardly anyone will take this into account. Obviously, a person buying a CPU of this class has slightly different priorities, and the determining factor is such productive characteristics of the processor as the ability to produce and process the necessary content.

We will end with a summary of the performance comparisons of the 18-core Core i9-7980X under various workloads.

We originally compiled it for our Threadripper chip review, and in our opinion, it's a great way to get a sense of what you can expect from these processors in reality. Compared with just the 10-core Core i9-7900X versus the 16-core Threadripper 1950X, the Core i9 pulled ahead under light workloads, but the AMD processor led the way in heavy tasks.

With the advent of the new Core i9, the situation is completely different. Now Intel products take the lead not only in light tasks, but also in the heaviest load are not inferior to the championship. If you look at the Cinebench R15 results below, you can see that the 18-core from Intel is not inferior to the chip from AMD by an inch.

Using CineBench R15, we changed the processor load from one thread to 36 - just to demonstrate the peaks in performance

Intel i9 price - if you really want to know

The question mark that looms over the Core i9 and the entire Core X series is the price quote. Ever since we released our first reviews of the Core i9-7900X and Threadripper 1950X, we were pretty confident that Intel would ultimately lead the way in performance without question.

The problem is that his products are also leading in terms of prices. Attempting to set a value based on performance is a slippery slope because the value of performance is relative. We've just seen that in general the Threadripper is only slightly below the speed of the Core i9. Therefore, we decided to line up all Core X and Threadripper processors not at the price of the chip itself, but at the "cost per thread." We even included the 10-core Core i7-6950X on this list, at a retail price of under two thousand dollars - that's just for fun.

Why isn't President Ben Franklin smiling? He probably just paid $ 1,723 for a Core i7-6950X Broadwell-E

Stream by stream, the worst value is, of course, the Broadwell-E chip. As expected, Intel's Core i5-7640X also came in second from the end. But the champion in terms of price-quality ratio, surprisingly, is AMD's development: 16-core and 32-thread Threadripper 1950X.

Conclusion

So, there are two ways to evaluate the Core i9. The first one is from the point of view of performance, where there are no questions at all, who is the champion here. You will have to look at the diagrams for a very long time and carefully to notice which of the multi-threaded tasks of the 16- and 18-cores Core i9 was able to bypass AMD's Threadripper. And if you move on to the lighter tasks that Intel's high-speed designs crack like nuts, things become even clearer.

So for performance maniacs who absolutely, absolutely, desperately need the fastest processors for tasks of any level of complexity, both chips, Core i9-7960X and Core i9-7980X, are new high-speed demons, a processor-dream.

The problem, of course, is the price difference. Our final chart just above may give you some thoughts on the value of AMD's offering. Yes, the Core i9 may be the officially recognized speed leader in every measure that can be measured, but it can't beat its own price tag.

Maybe it depends on who is paying. If, for example, your boss instructs you to find a new workhorse for editing video files, you will probably lean in favor of Intel. But if you collect this car with your own earned kopecks and try to stretch every ruble up and down? AMD might be a natural choice in this case.

And yet, make no mistake. Core i9 today is a clear and undisputed performance leader.

The Core i9 lineup is something new on Intel's mainstream processor lineup, with its lineup until recently consisted of Celeron, Pentium, Core i3, i5 and i7. First, in order to highlight the new models, it was necessary to increase the number designation. Secondly, in order to indicate changes in the kernel, it is more profitable to put new solutions in a single group with a high index.

advertising

Both factors are key for Intel, but not at all because AMD is stepping on the heels with its Ryzen and Threadripper, but because the market is really waiting for something new.

The Intel Core i9-7900X processor, which is already presented with the advanced architecture of Kaby Lake, does not fall due to a step back for the CPUs released by Intel for enthusiasts, we are already used to it.

But in order not to thicken the clouds over the fastest system on the X299 chipset, quad-core Kaby Lake-X were added to Skylake-X. In Skylake-X themselves, and they, as before, are part of Intel's server strategy, caches were rebalanced, the ring bus was replaced, and other changes.

It is worth starting with the reasons for the rejection of the ring bus. Hereinafter, we will speak on behalf of Intel, giving an answer to the question "Why?" It's no secret that the company has been increasing the number of cores in server processors for a long time, and sooner or later the ring topology would become a bottleneck. But it was difficult to take and refuse a fast "carousel" for the sake of something new. When designing, we took into account the features of other options and came to a matrix system.

advertising

This complicated the task for engineers and added more than one extra million transistors to the processor, all for the sake of the desire to put more cores on one die. Moreover, if you visualized how such a system works, then to understand it, let us explain that at each point-stop there is now both an interface and a router that manages data. In other words, the matrix will potentially increase the number of cores, and the more there are, the more profitable the use of the new communication system becomes.

Behind the scenes, the question of the percentage of yield of suitable crystals with a constant increase in area remains. Wouldn't a single global hub approach be more profitable and cheaper than using a matrix bus?

The fastest processor for enthusiasts on the Broadwell architecture carried ten cores, and the server version was equipped with as many as twenty-two. Let me remind you that it was the Broadwell core that was the fastest in terms of one megahertz frequency. Naturally, the Intel Core i7-6950X turned out to be very fast, but at the same time it was poorly overclocked. The average overclocking was in the range of 4.0-4.2 GHz.

In his defense, I will say that this was enough to solve a wide range of tasks. At the same time, the Core i7-6950X could not be called a gluttonous or very hot processor. The second negative factor is the price. Not every rich user could master the purchase of a top-end Broadwell representative at a cost of $ 1723. So if there was a demand, it was exclusively from the loyal fans.

The Skylake-X processors that replaced Broadwell-E did not differ much in their aggregate characteristics: the number of cores remained the same, the frequency increased from 3.0 / 3.5 / 4.0 GHz to 3.3 / 4.0 / 4.5 GHz. But at the same time, the volume of the L2 cache memory increased from 2.5 MB to 10 MB (and it became almost twice as slow), the volume of the L3 cache, on the contrary, decreased from 25 MB to 13.75 MB (and it itself became 40 % faster), added support for AVX512.

So far, only solutions with six, eight and ten cores will be available. Theoretically, in the future, for the same money ($ 1723), we will get a sixteen-core processor, but when this will happen is unknown.

Specifications

Model	Clock frequency, GHz	Clock frequency, GHz (Turbo)	Number cores	Number streams	Cache memory L1, MB	L2 cache, MB	Cache memory L3, MB	Maxi- minor calculated power, W	Recommendations given cost, $
AMD Ryzen 7 1800X	3.6	4.0	8	16	0.7	4	16	95	399
Intel Core i9-7900X	3.3	4.3	10	20	0.6	10	13.75	140	989
Intel Core i7-6950X	3.0	3.5	10	20	0.6	2.5	25	140	1 723

Test stand

Test configuration # 1 (Intel Kaby Lake-X / Skylake-X)

• Motherboard: ASUS Prime X299-Deluxe (Intel X299, LGA 2066);
• RAM:
• DDR4 Corsair Vengeance LPX, 4 x 4 GB, 2800 MHz 16-18-18-36-2T;
• DDR4 G.Skill F4-3600C17D, 2 x 4 GB, 2133 MHz 17-18-18-38-1T @ 3333 MHz 17-18-18-38-1T;
• Drives:
• SSHD Seagate Desktop 4 TB;

• Intel Core i9-7900X 3.3 GHz, Turbo Boost up to 4.5 GHz, ten cores, twenty threads;
• Intel Core i7-7740X 4.3 GHz, Turbo Boost up to 4.5 GHz, four cores, eight threads;
• Intel Core i9-7900X @ 4.0 GHz, 40 x 100 MHz, ten cores, twenty threads;
• Intel Core i9-7900X @ 4.5 GHz, 45 x 100 MHz, ten cores, twenty threads;
• Intel Core i7-7740X @ 4.5 GHz, 45 x 100 MHz, four cores, eight threads.

Test configuration # 2 (Intel Kaby Lake / Skylake)

• Motherboard: ASUS Maximus IX Formula (Intel Z270, LGA 1151);
• Cooling system: water cooling system;
• Thermal interface: Arctic Cooling MX-2;
• RAM: DDR4 G.Skill F4-3600C17D, 2 x 4 GB, 2133 MHz 17-18-18-38-1T @ 3333 MHz 17-18-18-38-1T;
• Video card: Nvidia GeForce GTX 1060;
• Drives:
• SSD Samsung 840 Evo, 240 GB;
• SSHD Seagate Desktop 4 TB;
• Power supply: Corsair AX1500i, 1500 watts;
• Operating system: Microsoft Windows 10 x64.

advertising

Processors and their modes of operation:

• Intel Core i7-7700K 4.2 GHz, Turbo Boost up to 4.5 GHz, four cores, eight threads;
• Intel Core i5-7600K 3.8 GHz, Turbo Boost up to 4.2 GHz, four cores, four threads;
• Intel Core i7-6700K 4.0 GHz, Turbo Boost up to 4.2 GHz, four cores, eight threads;
• Intel Core i7-7700K @ 4.5 GHz, 45 x 100 MHz, four cores, eight threads;
• Intel Core i5-7600K @ 4.5 GHz, 45 x 100 MHz, four cores, four threads;
• Intel Core i7-6700K @ 4.5 GHz, 45 x 100 MHz, four cores, eight threads.

Test configuration # 3 (Intel Broadwell-E)

• Motherboard: ASUS X99-Deluxe II (Intel X99, LGA 2011-3);
• Cooling system: water cooling system;
• Thermal interface: Arctic Cooling MX-2;
• RAM: DDR4 Corsair Vengeance LPX, 4 x 4 GB, 2800 MHz 16-18-18-36-2T;
• Video card: Nvidia GeForce GTX 1060;
• Drives:
• SSD Samsung 840 Evo, 240 GB;
• SSHD Seagate Desktop 4 TB;
• Power supply: Corsair AX1500i, 1500 watts;
• Operating system: Microsoft Windows 10 x64.

Processors and their modes of operation:

• Core i7-6950X 3.0 GHz, Turbo Boost up to 4.0 GHz, ten cores, twenty threads;
• Core i7-6950X @ 4.0 GHz, 40 x 100 MHz, ten cores, twenty threads.

Test Set # 4 (AMD Ryzen)

• Motherboard: ASUS ROG Crosshair VI Hero (AMD X370, Socket AM4);
• Cooling system: water cooling system;
• Thermal interface: Arctic Cooling MX-2;
• RAM: DDR4 Geil Evo X, 2 x 8 GB, 2133 MHz 17-18-18-38-1T @ 3200 MHz 17-18-18-38-1T;
• Video card: Nvidia GeForce GTX 1060;
• Drives:
• SSD Samsung 840 Evo, 240 GB;
• SSHD Seagate Desktop 4 TB;
• Power supply: Corsair AX1500i, 1500 watts;
• Operating system: Microsoft Windows 10 x64.

advertising

Processors and their modes of operation:

• AMD Ryzen 7 1800X 3.6 GHz, Turbo Boost up to 4.0 GHz, eight cores, sixteen threads;
• AMD Ryzen 7 1800X @ 4.0 GHz, 40 x 100 MHz, eight cores, sixteen threads;
• AMD Ryzen 5 1600X 3.6 GHz, Turbo Boost up to 4.0 GHz, six cores, twelve threads;
• AMD Ryzen 5 1600X @ 4.0 GHz, 40 x 100 MHz, six cores, twelve threads;
• AMD Ryzen 5 1400 3.6 GHz, Turbo Boost up to 4.0 GHz, four cores, eight threads;
• AMD Ryzen 5 1400 @ 3.9 GHz, 39 x 100 MHz, four cores, eight threads.

Memory frequency and timings

Intel Core i7-7700K @ 4.5	3333 MHz 17-18-18-38-1T
Intel Core i7-7700K	2133 MHz 17-18-18-38-1T
Intel Core i5-7600K @ 4.5	3333 MHz 17-18-18-38-1T
Intel Core i5-7600K	2133 MHz 17-18-18-38-1T
Intel Core i7-6700K @ 4.5	3333 MHz 17-18-18-38-1T
Intel Core i7-6700K	2133 MHz 17-18-18-38-1T
AMD Ryzen 7 1800X @ 4.0	3200 MHz 17-17-17-37-1T
AMD Ryzen 7 1800X	2133 MHz 17-17-17-37-1T
AMD Ryzen 5 1600X @ 4.0	3200 MHz 17-17-17-37-1T
AMD Ryzen 5 1600X	2133 MHz 17-17-17-37-1T
AMD Ryzen 5 1400 @ 3.9	3200 MHz 17-17-17-37-1T
AMD Ryzen 5 1400	2133 MHz 17-17-17-37-1T
Intel Core i7-6950X @ 4.0	2800 16-18-18-36-2T
Intel Core i7-6950X	2800 16-18-18-36-2T
Intel Core i7-7740X @ 4.5	3333 MHz 17-18-18-38-1T
Intel Core i7-7740X	2133 MHz 17-18-18-38-1T
Intel Core i9-7900X @ 4.5	2800 16-18-18-36-2T
Intel Core i9-7900X @ 4.0	2800 16-18-18-36-2T
Intel Core i9-7900X	2800 16-18-18-36-2T

Intel's HEDT platform update has been planned for a long time. A year ago, when the company released its Broadwell-E processors, it was known that they were coming only for a year and this summer they should be replaced by the newer Skylake-X. However, nothing particularly interesting was expected from this event. Notable in the planned announcement was perhaps only the fact that Intel was going to close the existing architectural gap between mass and high-performance chips and release within the new version of the HEDT platform not only CPUs based on the Skylake design (which was presented back in the summer of 2015), but and chips with the most recent Kaby Lake architecture. However, multi-core processors for desktop systems were to be released only in the Skylake-X family, and the Kaby Lake-X family was to include only additional and secondary quad-core chips, which are essentially analogs of the mass Kaby Lake for the LGA1151 platform.

Thus, from the point of view of enthusiasts, the HEDT platform should have continued its systematic movement in its usual course: a little more cores, a little higher frequency, a slightly different socket, slightly increased prices, etc. And we have no doubt that everything would be so it would have been if Ryzen hadn't happened this spring. The new architecture presented by AMD turned out to be so successful, and the pricing policy of this company turned out to be so daring that Intel simply could not leave the inclinations of a competitor without any response. Moreover, AMD also announced the Threadripper project, in which the intention was to encroach on the holy of holies - a segment of high-performance platforms with multi-core processors, where Intel has long considered itself the only and unique player.

As a result, the new Skylake-X processors we are talking about today received two fundamentally important unexpected changes.

First, Intel decided not to restrain itself in increasing the number of processor cores, and desktop CPUs with 12, 14, 16 and 18 cores are expected within the new platform. This means that for the first time Intel will offer enthusiasts not only adapted versions of Skylake-SP server processors based on the simplest version of the semiconductor crystal LCC (Low Core Count), but also processors on a chip of medium complexity HCC (High Core Count), which will allow more confidently address the HEDT platform to an audience of professionals - video content creators, modelers and developers working with ultra-high resolutions and virtual reality.

The second change is even more striking and concerns pricing. Skylake-X processors have become significantly cheaper than their predecessors. If in the Broadwell-E family a ten-core processor cost $ 1,723, then a Skylake-X similar in number of cores will cost only $ 999. Similar changes apply to the rest of the lineup. In general, if earlier prices for senior HEDT-class processors were formed according to the principle of "$ 170 per core", now for multi-core Skylake-X there will be a much more liberal rule "$ 100 per core".

Ultimately, the new incarnation of the HEDT platform becomes more accessible and closer to the end user. The number of scenarios where this platform can be used increases, and the entry threshold decreases. In other words, Skylake-X and Kaby Lake-X processors no longer seem so elite and status products. Obviously, the number of those who want to buy them, and not the flagship LGA1151 chips, will obviously be greater than before. And in this review, we will take a closer look at the new HEDT platform and the ten-core Core i9-7900X processor - the older version of Skylake-X for the next couple of months, which will appear on store shelves in a week.

⇡ Skylake-X processors: general information

Codenamed Basin Falls, Intel's new HEDT platform is a much more comprehensive and scalable product than the high-performance platforms of previous generations that used LGA2011 and LGA2011-3 processor sockets.

Previously, the lineup in each generation of the HEDT platform included only three or four CPUs, the number of cores of which differed by no more than one and a half to two times. Now there will be at least nine processors compatible with the Basin Falls platform, and the difference in the number of cores between the simplest and the most sophisticated chip will be more than four times. Against this background, it is not at all surprising that the new HEDT processors are divided into three groups, differing in design and architecture, but compatible with the same LGA2066 processor socket.

	Kernels / threads	Base frequency, GHz	Turbo mode, GHz	Turbo Boost Max 3.0, GHz	L3 cache, MB	PCI Express 3.0 Lines	Memory channels	Memory frequency	TDP, W	Price
Skylake-X (HCC)
Core i9-7980XE	18/36	?	?	?	?	44	?	?	?	$1999
Core i9-7960X	16/32	?	?	?	?	44	?	?	?	$1699
Core i9-7940X	14/28	?	?	?	?	44	?	?	?	$1399
Skylake-X (LCC)
Core i9-7920X	12/24	?	?	?	?	44	?	?	?	$1199
Core i9-7900X	10/20	3,3	4,3	4,5	13,75	44	4	DDR4-2666	140	$999
Core i7-7820X	8/16	3,6	4,3	4,5	11	28	4	DDR4-2666	140	$599
Core i7-7800X	6/12	3,5	4,0	No	8,25	28	4	DDR4-2400	140	$389
Kaby lake-x
Core i7-7740X	4/8	4,3	4,5	No	8	16	2	DDR4-2666	112	$339
Core i5-7640X	4/4	4,0	4,2	No	6	16	2	DDR4-2666	112	$242

A couple of the simpler chips, Core i7-7740X and Core i5-7640X, have four cores with or without Hyper-Threading technology and belong to the Kaby Lake-X class. They are 100-200 MHz faster analogs of the Core i7-7700K and Core i5-7600K, ported to another socket. There is no difference in architecture and in specific performance, however, due to a more liberal thermal package, a tightly locked graphics core and changes in the power scheme, some improvements may occur in the overclocking potential.

We will take a closer look at the properties of representatives of the Kaby Lake-X series in one of the following reviews, since their sale should begin simultaneously with Skylake-X in the very near future. However, it should be borne in mind that due to the peculiarities of its origin, Kaby Lake-X seem to be frankly flawed proposals against the background of Skylake-X, not only because of the small number of cores. They also use a simplified dual-channel memory controller and a PCI Express controller that only supports sixteen lanes. This means that, although Kaby Lake-X is designed to operate as part of the Basin Falls platform, they will not provide a significant part of its key benefits.

Skylake-X processors are of much greater interest to high-performance enthusiasts: they allow using all the capabilities of the Basin Falls platform to the fullest and can be considered as full-fledged heirs of the previous generation of HEDT chips, Broadwell-E. However, in the Skylake-X generation, Intel's approach under the influence of the active actions of a competitor has undergone some changes, and new items belonging to this class were divided into two groups: processors with a relatively small number of cores and processors - multi-core monsters.

The standard strategy that the microprocessor giant has always used when creating consumer chips for the upper market segment has been to adapt server processor variants with a relatively small number of cores based on semiconductor crystals of LCC for such needs. And this strategy has worked successfully over the past several years. So, server processors are traditionally divided into three classes, for each of which their own semiconductor crystal design is developed: LCC (Low Core Count), HCC (High Core Count) and XCC (Extreme Core Count). In the Broadwell-EP generation, the first class included chips with up to ten cores, respectively, the older consumer LGA2011-3 CPUs are ten-cores. In the Skylake-SP generation, the LCC has already received twelve cores. And it is quite natural that the Skylake-X processors, which were originally planned for the Basin Falls platform, should have received from six to twelve cores.

Thus, all Skylake-X with the number of cores from six to twelve and support for Hyper-Threading technology are completely traditional high-performance chips for desktop computers. They are based on the same 14nm 12-core semiconductor LCC with the Skylake microarchitecture, in which up to six cores can be disabled to form certain CPU models. In addition, differentiation among such processors occurs in the number of PCI Express lanes supported by the controller built into the CPU. The older models with ten and twelve cores offer 44 PCI Express lanes, while for processors with six and eight cores, the PCI Express controller only supports 28 lanes.

LCC crystal: 12 cores, area 325 mm 2

But all the LLC-based Skylake-X variants have comparatively high clock frequencies. The thermal package of such processors is set at 140 W typical for the HEDT platform, but their frequencies are significantly increased compared to Broadwell-E. The ten-core Core i9-7900X has a 3.3 GHz base frequency and can be turbocharged up to 4.3 GHz; the base frequency of the eight-core Core i7-7820X is set at 3.6 GHz with a similar turbo mode at 4.3 GHz, and the nominal frequency of the six-core Core i7-7800X is 3.5 GHz with the ability to automatically overclock under low load to 4.0 GHz. The full passport characteristics of the twelve-core Core i9-7920X have not yet been announced - this processor should be released only in a couple of months.

It is worth paying attention to one more interesting point. With the advent of the Basin Falls platform, Intel's Core i9 processors appear in the range. Thus, Intel decided to emphasize the elitism of individual Skylake-X models, which, most likely, will be directly opposed to AMD Threadripper. But for now, the principle of naming the Core i9 is purely formal. It is received by processors with more than 10 cores and 44 PCI Express lanes. This means that before the 12-core processor, scheduled for August, there will be only one Core i9 in the Skylake-X line - the ten-core thousand-dollar Core i9-7900X.

But by the way, it's not a fact that with the release of the 12-core Core i9-7920X, the current sub-flagship Core i9-7900X will fade against its background. Intel's failure to release its twelve-core processor along with the rest of the LLC's Skylake-X processors is due to the fact that the company cannot yet decide whether to make it more economical or faster. In theory, the LGA2066 platform supports processors with a typical heat dissipation of up to 165W, which allows you to set the Core i9-7920X frequencies to a high enough level, but Intel does not want to resort to this measure to avoid the incompatibility problems with motherboards and cooling systems, which can certainly arise from - because the company has not yet released such hot processors. Therefore, it was decided to take a pause, during which Intel engineers hope to understand how impressive AMD's HEDT platform will turn out.

In addition, Intel has another powerful tool that it can oppose to AMD HEDT processors - Skylake-X chips based on the HCC crystal. This crystal has 18 cores and in the future will release three additional versions of the Core i9 with 14, 16 and 18 cores. For obvious reasons, the exact characteristics of these models have not yet been determined, and their release is scheduled only for October. However, Intel already wants to secure the title of the manufacturer of HEDT processors with the largest number of cores, leaving some room for maneuver with frequencies and heat dissipation.

HCC crystal: 18 cores, area 484 mm 2

Ultimately, the Basin Falls platform looks like a noticeable step forward. Skylake-X has received an impressive and versatile set of improvements over Broadwell-E. Starting with the fact that the new processors offer a significantly increased number of cores and noticeably increased operating frequencies, and they do this with a passing price reduction. And ending with the fact that Skylake-X implements a more powerful four-channel memory controller with official support for DDR4-2666, as well as a PCI Express 3.0 controller with an increased number of lines by four pieces. Along the way, do not forget about the new Skylake microarchitecture, which itself contains a number of optimizations that allow you to raise the specific performance at a constant frequency.

And here one more important detail must be emphasized. The microarchitecture of the cores of the new Skylake-X processors does not just repeat the familiar Skylake microarchitecture of 2015. Additional improvements have been added to new HEDT products, which we will discuss in detail below. Among them: support for 512-bit vector instructions AVX-512, change the cache memory subsystem, change the topology of inter-core connections and a new version of Turbo Boost Max 3.0 technology, which allows you to raise the frequencies of a selected pair of processor cores to 4.5 GHz.

⇡ Set of system logic Intel X299 and LGA2066-motherboards

Along with the new Skylake-X and Kaby Lake-X processors, Intel is bringing to the market the counterpart of the Basin Falls platform - a new set of system logic X299. However, we would not argue that this chipset is as innovative as the accompanying processors. In a nutshell, it should be said that the X299 brings to the HEDT platform only those features that have long become standard for LGA1151 systems. However, such a change should not be underestimated. Chipsets for LGA2011 and LGA2011-3 systems were much less functional. And if the X299 is compared with the X99, and not with the Z270, then the progress becomes obvious.

There are two main changes. Firstly, the X299 has a standard HSIO topology (High-Speed \u200b\u200bIO). This means that the new set of logic is like a PCIe switch: it has 30 high-speed ports that motherboard manufacturers can flexibly configure to suit their needs and ultimately receive the required number of PCI Express 3.0 lanes, as well as USB 3.0- and SATA 3.0- ports. Secondly, the bus used by the chipset to communicate with the processor has changed. If the X99 used the DMI 2.0 bus for these purposes, then the X299 switched to a twice as fast DMI 3.0 bus, in many respects similar to PCI Express 3.0 x4.

The high-speed ports of the chipset allow you to get from it in different combinations up to 24 PCI Express 3.0 lanes, up to eight SATA 3.0 ports and up to ten USB 3.0 ports. This is almost equivalent to the capabilities of the Z270, and one would think that the X299 hub is a variation of the logic set from the LGA1151 platform, but the X299 still has a unique feature - it supports a couple more SATA ports. The rest of the characteristics are similar. Moreover, this also applies to the fact that both chipsets are manufactured using the same 22-nm process technology, have the same heat dissipation at the level of 6 W, and even differ little from each other externally.

Honestly, from X299, which comes along with the Basin Falls platform for a relatively long time, I would like some additional features, for example, support for USB 3.1 Gen 2 and WiFi, which should appear in the next generation of logic sets for the LGA1151 platform. But there is nothing like that in X299, and all such functions are left to the mercy of motherboard manufacturers, who will again be forced to complete their flagship LGA2066 solutions with a scattering of additional controllers.

On the other hand, the X299 supports Intel Optane drives and all other functions implemented via the Intel RST 15 driver. This, in particular, means that PCIe drives connected to the chipset can be used to form RAID arrays of levels 0, 1, and 5. Moreover the number of participants in such arrays can be up to three.

However, given the rich set of PCI Express lanes available on the processor, motherboard manufacturers will most likely implement M.2 slots connected directly to the CPU. Especially for such cases, the Basin Falls platform has an additional unique function VROC (Virtual RAID On CPU). It allows you to combine any number of PCI Express drives connected directly to the processor into RAID arrays. True, this technology has some offensive software limitations. For example, to activate RAID modes other than RAID 0, the user will need a special key, which will need to be purchased separately.

Along with the new set of logic, the Skylake-X and Kaby Lake-X processors also require a new 2066-pin LGA2066 (Socket R4) connector. The need to implement a new socket in this case was due to the transition to DMI 3.0 and the appearance of several additional PCI Express lines in the processor, so there is no and cannot be compatibility between the new HEDT processors and previous platforms with the LGA2011-3 socket.

Nevertheless, in appearance and dimensions, the LGA2066 does not differ much from the LGA 2011-3. Moreover, Intel has managed to maintain full compatibility with older cooling systems. The way of attaching the coolers to the socket remains the same as before, the location of the mounting holes has not changed either. Accordingly, old Haswell-E and Broadell-E coolers will fit the new Skalake-X and Kaby Lake-X processors without any restrictions.

Because the Kaby Lake-X and Skylake-X processors vary greatly in performance, including the number of PCI Express processor lanes and the number of memory channels, the LGA2066 platform has flexibility that has not yet been found. Intel's LGA2066 motherboard requirement requires all motherboards to support the full line of LGA 2066 processors, with no exceptions. This means that a typical LGA2066-board should allow building configurations with both dual-channel and four-channel memory subsystems, as well as with 16, 28 or 44 PCI Express lanes coming from the CPU.

And this is actually not an easy task, the solution of which leads to the fact that buyers of inexpensive LGA2066 processors will have to pay extra for features that they will most likely never use. Although we do not exclude the possibility that motherboards optimized for younger LGA2066 processors and with a reduced number of DIMM and PCI Express slots may appear on sale, in most cases the situation will most likely be such that when installing Kaby Lake-X, some of the slots the motherboard will be rendered unavailable for use.

Something similar will happen when installing Kaby Lake-X and lower versions of Skalake-X not only with DIMM slots, but also with PCI Express processor slots. Some of them can be switched off, while the other part can switch to "weaker" speed modes.

⇡ New in Skylake-X

New architecture of cache memory

Skylake-X processors should not be seen as a simple transfer of the familiar Skylake microarchitecture to a multi-core design. Over the past two years since its inception, Intel engineers have done some work and made some changes to the original project. Therefore, Skylake-X processors can be considered carriers of an updated version of the basic microarchitecture, which ultimately endows them with slightly different specific performance (in terms of frequency). And the most important improvement concerns the alteration of the cache memory subsystem in order to increase its efficiency.

In HEDT processors of previous generations (as well as in Xeon), the cache memory architecture assumed the allocation of its own L1 and L2 caches for each core and the presence of a single L3 cache for all cores, which was inclusive and had an impressive size. This meant that all the data that was in the L2 cache was duplicated in L3, however, if data from the L2 cache was preempted, it was still available in L3. Such a scheme of work was quite profitable, and its efficiency was largely supported by a correctly selected ratio between the volumes of cache memory of different levels. While the L2 cache had a capacity of 256 Kbytes, the L3 cache size was formed from 1.5 to 2.5 Mbytes per core. As a result, despite the costly inclusive algorithm, L3 retained enough space to independently work with data.

However, in Skylake-X, it was decided to change the balance. Considering that L2 cache has much better latency rates, and its capacity has a stronger effect on performance, it was decided to increase its size in new processors to 1 MB, that is, four times. At the same time, in order not to go beyond the acceptable transistor budget, this was done simultaneously with a decrease in the L3 cache shared between the cores, the volume of which in Skylake-X is now determined at the rate of 1.375 MB per core.

Along the way, in order to preserve the efficiency of the L3 cache with a serious decrease in volume, the algorithm of its functioning was changed. Now this cache is not inclusive, and moreover, it is victimized. This means that the L3 cache is filled exclusively by pushing data out of L2, and the data prefetching mechanisms do not apply to it. Ultimately, this means that while the effective total cache size for the Haswell-E and Broadwell-E processors was 2.5 MB per core, for Skylake-X it remained almost the same - 2.375 MB per core. However, the Skylake-X caching system should provide lower latencies on average, since a significant part of the cache memory is of the second level, which is characterized by low latency.

The Skylake-X cache memory structure is described in more detail in the table:

At the same time, the L3 cache of Skylake-X processors has clearly become worse in terms of the algorithm of work, and in terms of associativity (that is, in terms of efficiency), and in terms of volume, and even in terms of operating frequency. However, all this, in the opinion of Intel engineers, should be compensated for by a more spacious L2 cache with twice as high associativity. According to the calculations presented by the developers, expanding the size of the L2 cache four times doubles the probability of finding the data necessary for the processor. This, in turn, reduces the downtime of the executive pipeline and, according to Intel engineers, increases the specific productivity by an additional 5-10 percent. Thus, thanks to changes in the cache memory subsystem, Skylake-X processors should outperform the usual Skylake-S and Kaby Lake-S even on a single-threaded load.

However, before taking such statements on faith, let's see how matters stand with the real latency of the cache memory subsystem in Broadwell-E and Skylake-X processors. To do this, using the SiSoft Sandra test suite, we measured the real latency when the processors access data blocks of various sizes. Both processors participating in the test worked at the same 4 GHz frequency and were equipped with a four-channel DDR4-3000 SDRAM with CAS Latency 15.

Frankly speaking, the situation with the real latency of the Skylake-X cache memory subsystem does not look very encouraging. Older Broadwell-E processors almost always provide lower data access times, except for the case when they do not fit into the L2 cache, but fit into it in Skylake-X. Therefore, the correctness of Intel's claims can be questioned. It seems somewhat implausible that the demonstrated latency gain will be sufficient for Skylake-X to gain any performance advantage in real-world applications.

However, in fairness, it is worth noting the higher practical bandwidth of the Skylake-X cache memory subsystem, which can serve as some compensation in a situation with delays.

Particularly pleasing against the background of high latency is the throughput of the L3 cache. Along with the revision of its architecture, Intel engineers were able to achieve significant increases in bandwidth. Why this happened will become clear from the next section.

⇡ Changes in the topology of inter-core connections

Along with the change in the caching system, Intel has completely redesigned the scheme that is used to organize inter-core communication. Recall that since the days of Sandy Bridge, a bidirectional 256-bit ring bus based on the QPI protocol has been used to connect processor cores and exchange data with the L3 cache and memory controller in Intel processors. And as long as the processors did not contain too many cores, this approach was very effective. A fairly simple circuit design really made it possible to achieve data transmission with minimal delays.

However, as the number of cores grew, data paths began to lengthen, and this began to cause serious problems. To ensure the smooth operation of multi-core Intel processors, they even had to switch to a scheme with the division of cores into two clusters and the introduction of two ring buses connected by two buffering bridges. But such a combination of cores, memory controllers and I / O controllers inside the processor could no longer boast of its former efficiency. If there was a need to transfer data between points located in different clusters, latencies suffered greatly. And ultimately Intel came to a situation where the ring bus became an obstacle to increasing throughput and reducing latency during intra-processor data operations.

Therefore, in the Skylake-SP server processors (and related HEDT Skylake-X processors), where the number of cores can reach 28 pieces, Intel switched to a different scheme of inter-core connections - a mesh network, which has already been well-tested in Intel Xeon Phi (Knights Landing ). The number of compounds in it is much larger, since all the cores on the chip are penetrated by through horizontal and vertical links. But due to this, the routes required for communication between cores and other functional nodes are significantly simplified, reducing latencies and equalizing the delays that arise during various interactions within such a network. In addition, such a network provides a higher total throughput.

This change allows the frequency of this network to be set below the frequency of the ring bus while maintaining high throughput rates. This means that the new mesh structure of the connection is not only good in balance and scalability, but also benefits in terms of resource consumption.

Naturally, all this is important primarily for server processors with a large number of cores, but Skylake-X turned out to be hostages of the situation: in them, the mesh network also replaced the ring bus. And in relatively simple cases, when the number of nuclei is not so large, the latencies in internuclear interaction have worsened compared to Broadwell-E. To test, we measured the latencies that occur when transferring data from one core to another for the ten-core Broadwell-E and Skylake-X. For the purity of the experiment, both processors worked at the same 4.0 GHz frequency.

As you can see from the illustration, Skylake-X's latency during internuclear communication is about 1.5 times higher. And this clearly means that the mesh network does not give any gain in the case of ten cores, but on the contrary, only worsens the situation.

A noticeable result of the changes that have taken place is changes in the speed of the memory subsystem. Since the DDR4 controllers in Intel processors are connected to the cores via the same bus as the cores to each other, the speed of the memory subsystem is directly related to the efficiency of the inter-core circuitry.

Using the Cachemem test from the AIDA64 package, we measured the performance of a memory subsystem made up of four identical DDR4-3000 SDRAM modules for Broadwell-E and Skylake-X processors operating at the same 4.0 GHz frequency, and the diagnosis was confirmed. The latencies inside the new generation chips have indeed become higher.

True, in fairness, it is worth noting the fact that along with the latency, the practical throughput when reading from memory has grown, which can compensate for the increased delays during streaming operations with large amounts of data. However, this consolation is rather weak, since in real tasks the latency of the memory subsystem has a very serious impact on performance.

⇡ Support for AVX-512 instructions

Speaking about what changes in the Skylake microarchitecture are timed to coincide with the release of the high-performance Skylake-X processors, we must mention that they have support for the new set of vector instructions AVX-512. It was first implemented in the latest generation of Xeon Phi (Knights Landing) computing accelerators, and now its support has reached traditional processors for servers, workstations and high-performance desktops.

In fact, the AVX-512 set is an extension of vector instructions for operations with 512-bit vectors. It has new 512-bit registers, new packed formats for integers and fractional numbers, as well as various operations on them. An important feature of the AVX-512 mode is the high speed of their execution: it is assumed that the processor can switch from ordinary 256-bit AVX instructions to 512-bit operations without slowing down the performance. And this fact allows Intel to present the promising 18-core processor as the first desktop processor with a performance of 1 teraflops.

In other words, the introduction of AVX-512 allows you to double the performance, but we are talking here exclusively about vector operations. If optimized for new instructions, parallel algorithms can actually be executed on Skylake about twice as fast, but this, of course, does not apply to ordinary general-purpose computations. Nevertheless, Skylake-X processors are quite capable of invading the territory where previously only video cards were used in calculations.

It's worth noting that the addition of AVX-512 support in Skylake-X is not only a forward-looking enhancement. Some existing algorithms have the necessary optimizations now and are able to gain a performance advantage. These include, for example, the popular x264 encoder, in which the community introduced support for new teams early this year.

To assess how much AVX-512 instructions are able to increase the performance of computational algorithms in a case close to ideal, you can use the Processor Multimedia synthetic test from the SiSoft Sandra package. This simple benchmark measures the speed of building a Mandelbrot set using a variety of instruction sets. With its help, we compared the performance of ten-core Broadwell-E and Skylake-X, operating at the same frequency of 4.0 GHz.

As you can see from the results, the use of 512-bit vector instructions alone can speed up calculations by an amount from 20 to 85 percent. And if we add to this the other architectural improvements incorporated in Skylake-X, it turns out that in terms of specific performance this CPU can more than double the Broadwell-E.

⇡ Enhanced Intel Turbo Boost Max Technology 3.0

With the release of the Broadwell-E processors, Intel introduced Turbo Boost Max 3.0 technology, which exploits the fact that the cores in a multi-core processor with a relatively large semiconductor crystal can vary significantly in their frequency potential. The idea was that among the processor cores there is probably one that can operate at a higher frequency and at a lower voltage, so it is logical to execute a low-flow load on it.

Intel implemented this principle through a dedicated driver that ported single-threaded applications to such a preselected production core. Motherboard manufacturers had to use BIOS to implement the possibility of increasing the operating frequency of this single core by an additional several hundred megahertz relative to the values \u200b\u200bprovided for by the classic Turbo Boost 2.0 technology. As a result, Broadwell-E multi-core processors, which have relatively low nominal frequencies, were able to solve single-threaded tasks with good efficiency.

In Skylake-X, this idea was further developed. Now in the processor for a low-threaded load, two special cores are selected at once, which makes it possible to get higher performance when running two single-threaded applications at once or when working in applications that can use two cores simultaneously.

True, it had to pay for this with an allowable increase in frequency within Turbo Boost Max 3.0. If in Broadwell-E processors this technology could raise the frequency of a selected core by 200-500 MHz, then in Skylake-X the additional acceleration is limited to only 200 MHz.

However, this may also be due to the fact that in the new generation of HEDT processors the classic Turbo Boost 2.0 technology is also very aggressive, leaving not too much free space for Turbo Boost Max 3.0.

⇡ Core i9-7900X Details

For testing, Intel provided us with the currently eldest Skylake-X processor, the ten-core Core i9-7900X. Recall that its sales will begin in a week, and more powerful representatives of the series will appear only in August (12-core Skylake-X) or in October (14-, 16- and 18-core Skylake-X).

The appearance of the LGA2066 processor is slightly different from the usual outlines of the LGA2013-3 processors, but the difference is not cardinal. The shape and dimensions remained approximately the same, in fact, only the differently designed edges of the heat dissipating cover stand out noticeably.

However, now this cover is not soldered to the semiconductor crystal of the processor, but contacts it through thermal paste.

In the CPU-Z diagnostic utility, the new Core i9-7900X does not look entirely obvious.

Please note that the utility identifies this processor as Core i7-7900X, and this is not a bug in the program. This name is really hardcoded into the processor itself as an identification string. The fact is that Intel decided to use the Core i9 brand quite recently, and the engineering samples sent out to reviewers contain a variant of the name originally planned.

Otherwise, all the characteristics of the Core i7-7900X sample are fully consistent with how the production Core i9-7900X processors will look like. This, in particular, is evidenced by the serial core stepping - H0.

The situation with the real operating frequencies of the Core i9-7900X is as follows:

• With a typical multi-threaded load on all cores, the frequency is most often at 4.0 GHz.
• If the multi-threaded load is especially resource-intensive, for example, it uses AVX instructions, the frequency can drop down to 3.3-3.6 GHz.
• Under a single threaded load, the frequency can be increased to the promised 4.5 GHz under the influence of Turbo Boost Max 3.0 technology. However, such automatic overclocking is not always observed, and in some situations the frequency under such conditions reaches only 4.1 GHz.

The thermal regime of the processor operating in the nominal does not cause any questions, despite the replacement of the solder under the processor cover with a polymer thermal interface. When testing the Core i9-7900X in LinX 0.7.2 (and this version already has support for the new AVX-512 instructions) using the Noctua NH-U14S single-tower cooler, the maximum temperatures on the internal processor sensor reached only 74 degrees, while the maximum allowable temperature for Skylake-X, 105 degrees are considered.

All this suggests that Intel's thermal paste in Skylake-X works more efficiently than in LGA1151 processors. Either its composition has changed, or the role is played by a noticeably large area of \u200b\u200bthe semiconductor crystal, which for LLC is about 325 mm 2 (versus 122 mm 2 for the quad-core Skylake-S).

Compared to its predecessor, the deca-core Broadwell-E, the new Core i9-7900X clearly wins in performance.

	Core i7-6950X	Core i9-7900X
Codename	Broadwell-e	Skylake-X
Production technology	14 nm, FinFET	14 nm, FinFET
Kernels / threads	10/20	10/20
Hyper-Threading Technology	there is	there is
Base frequency, GHz	3,0	3,3
Maximum frequency in turbo mode, GHz	3,5	4,3
The maximum frequency of Turbo Boost Max 3.0, GHz	4,0	4,5
Unlocked multiplier	there is	there is
TDP, W	140	140
L2 cache, KB	10 × 256	10 × 1024
L3 cache, MB	25	13,75
PCI Express 3.0 Lines	40	44
DDR4 SDRAM support	Four channels DDR4-2400	Four channels DDR4-2666
Instruction set extensions	SSE4.1 / 4.2, AVX 2.0	SSE4.1 / 4.2, AVX 2.0, AVX-512
Packaging	LGA 2013-3	LGA 2066
Price	$1 723	$999

With the transition to a new architecture, operating frequencies increased by 10-30 percent (depending on the mode), at the official level, compatibility with DDR4-2666 SDRAM appeared, support for AVX-512 instructions was added, and the amount of L2 cache increased. Only the volume of the L3-cache turned out to be in the red, which almost halved. However, the most important change is indicated in the last line of the table: the ten-core now costs 42 percent less.

The list of competitors has not changed over the years: you can compare the new product either with a similarly powerful processor from AMD, or with models from its own line.

For example, the Intel Core i9-7960X chip, although it is made according to the same process technology and on the same architecture, cannot boast of the same high clock speed, and the test results will be in favor of the review hero.

New Intel Core i9-7940X

What alternative can AMD offer? For example, AMD Ryzen 7 2700X, which is also geared towards gamers or professional tasks. It can offer a completely different architecture, fewer cores for the same performance. But the price tag will also be 25% lower.

Power at a high level

The chip boasts 14 Skylake-X cores that are clocked at 3.1 GHz. Not the highest frequency is due precisely to the large number of cores, but support for Turbo Boost technology neutralizes this small drawback: during overclocking, the cores maintain a frequency of 4.3 GHz, simultaneously processing data from 28 threads at once.

Intel Core i9-7940X testing

The processor contains four-channel DDR4 RAM and a good amount of cache. The second level cache is 14 × 1024 KB, while the third level cache boasts 19 MB.

Perhaps in terms of its technical characteristics, the hero of the review lags behind its counterparts, but the Intel Core i9-7940X has a lower TDP - only 165 W versus 180 W for the same AMD.

High power consumption - like all i9s

The minimum power consumption is unpleasantly surprising - almost 85 W, although even processors assembled from the X line consume at 80 W. At maximum load, the picture changes slightly: the hero of the review consumes 212 W, while the similar Core i9-7900X and Core i9-7960X will require 240 and 235 W, respectively.

Intel Core i9-7940X pad

Excellent test results

At first glance, it seems that this processor is suitable for almost any task. Tests confirm this: the PCMark 8 benchmark estimates the work of the hero of the review at 3899 points, and the computation time was only 1.6 seconds. In the Application Benchmark 2017 test, the chip is rated at an excellent 316 points. In these tests, the processor confidently takes the lead, outperforming, if not its older brothers, then the analog from AMD.

But in Cinebench R15, TrueCrypt 7.1 AES-Twofish-Serpent and PovRay 3.7 RC3, the situation changes: Intel Core i9-7940X lags slightly behind competitors, but not critical.

In terms of scientific calculations, the hero of the review significantly lags behind other Intel models: only 260 points scored against 280-290.

Intel Core i9-7940X packaged

Should I buy this chip to work with content? We look at the results of the corresponding checks: when processing photos in Adobe Photoshop, the chip gains almost 230 points, remaining on par with its Core i9. When it comes to video creation, the i9-7960X and i9-7900X are slightly ahead (they have 236 and 223 points against 210).

Bottom line: great option. But not for everyone

Intel engineers managed to create a good processor, which, however, is not very versatile. It seems that all the data is there, but still Intel Core i9-7940X demonstrates not the highest test results. Maybe you should pay a little extra and get a reliable assistant both in games and in calculations?