Introduction

Following its high-end desktop platform launch with the Core X family, Intel is closing the year with the 8th generation Core "Coffee Lake" mainstream processor family. 2017 has been an exciting year in the world of processors with AMD's Ryzen offering becoming an unexpected success, restoring much-needed competition to the CPU market. AMD is able to compete with Intel at nearly every price point, from the $120-ish entry-level all the way up to the $1,000 high-end desktop. This spells trouble for Intel's mainstream-desktop platform, which is one of its main cash cows for high-margin processors, besides the notebook processor market.

The 8th generation Core processor family is based on the new "Coffee Lake" silicon, which is the company's fourth to be built on the 14 nanometer process (after Broadwell, Skylake, and Kaby Lake). It is proof that Intel's "tick-tock" product development cycle is off the rails, and the company can no longer launch a new process every other year. In the face of a reinvigorated AMD, there was only one direction in which Intel could have enhanced its mainstream-desktop processor lineup – core counts.



Increased core counts is the defining feature of the 8th generation Core family and its flagship, the Core i7-8700K covered in this review. For close to a decade, Intel sold quad-core desktop chips under the Core i7 and Core i5 brands, and dual-core ones under the Core i3 brand. The Core i7 and Core i5 brands now consist of six-core chips, and the Core i3 brand features quad-core chips. The L3 cache amounts have been proportionately increased. The Core i7 chips include 12 MB of it and have HyperThreading, while the Core i5 chips lack HT and have their L3 cache set to 9 MB. The Core i3 quad-core chips lack HT as well and feature 6 MB of L3 cache.

The "Coffee Lake" micro-architecture does not improve the CPU-core design from the two-year old "Skylake" architecture in any tangible way. Over the past two years, Intel has refined its existing 14 nm process at the physical level to increase clock speeds at minimal power and thermal costs. The new "Coffee Lake" silicon is built on the company's very latest 14 nm++ node.

Intel kept its first wave of Core "Coffee Lake" processors rather compact with only two SKUs per brand, one of which is unlocked (K), per brand. In this review, we're testing the top-dog Core i7-8700K, which succeeds the previous-generation i7-7700K. Intel is marketing this part as the "fastest processor you can buy for gaming PCs."

A Closer Look

The 8th generation Core processor family retail boxes feature darker, more striking colors, which helps you differentiate them from the older generations. The Core i7-8700K retail box only includes the processor, a case-badge, and the usual documentation. You will have to install your own separately purchased cooling solution.


With this generation, the biggest point of confusion has been the package. The 8th generation Core desktop processors bear the "LGA 1151" package markings and look like they'll work on older 100-series and 200-series chipset motherboards. They'll even physically fit on them since nobody at Intel bothered to put the key notches elsewhere. The chips, however, will not work on older motherboards. The machine won't even POST. The box clearly states that you need a 300-series chipset motherboard to use the processor. This is because the pin maps between Coffee Lake and older Kaby Lake/Skylake chips are different. More pins are allocated for power delivery; according to Intel to make up for the increased power requirements due to the six-core configurations.


For its mainstream-desktop processors, Intel has maintained a largely uniform package size for the past decade, dating all the way back to Core "Lynnfield" LGA 1156. The cooler mount-hole spacing hasn't changed. You will be spoiled for choice when choosing a compatible cooler; however, make sure it can cope with a 91-95W TDP. Certain low-profile coolers designed for 65W or 45W TDP chips are not recommended.

Architecture

The 8th generation Intel Core processors are based on the "Coffee Lake" micro-architecture. The CPU circuit design is essentially the same as with "Skylake," but the silicon is built on Intel's third iteration of the 14 nanometer silicon fab process, which the company refers to as 14 nm++. This node improves the ability for the chipmaker to dial up clock speeds at minimal power/thermal cost. While the quad-core "Kaby Lake" silicon was "Skylake" built on a refined 14 nm+ process, the six-core "Coffee Lake" silicon is a new design with a die-area of 150 mm².

The "Coffee Lake" silicon physically features six CPU cores, with 256 KB of dedicated L2 cache per core and 12 MB of shared L3 cache. The integrated Gen 9.5 graphics core is physically carried over from the "Kaby Lake" die, but is bolstered by higher clocks and an enhanced driver, which lets Intel brand it the "Intel UHD Graphics 600 series." Internal communication is handled by a "ring bus" and not the mesh-interconnect Intel deployed on its new Core X "Skylake-X" processors.



The system agent (the integrated Northbridge) also appears to be carried over from the "Kaby Lake" die, with its dual-channel DDR4 memory interface. There are minor improvements, such as the standard DDR4 memory clock being upped to DDR4-2666 on the Core i7 and Core i5 SKUs, and DDR4-2400 on the Core i3 SKUs. The IMC supports XMP 2.0 profiles. The processor only puts out 16 PCI-Express gen 3.0 lanes meant for PEG (PCI-Express discrete graphics). It talks to the motherboard chipset over the DMI 3.0 chipset bus, with a 32 Gbps-per-direction bandwidth.



The "Coffee Lake" CPU core is of the same exact design as Skylake and Kaby Lake, which dates back to 2015. Compared to the Haswell/Broadwell core, it features an improved front-end with a 25% fatter 5 µOP pipeline, a 50% wider allocation queue depth; an improved branch-prediction unit, and a wider instruction window. The execution stage features a slightly bigger re-order buffer, a bigger integer register file, an improved on-chip memory system. All of these contribute to a 5-10 percent IPC increase over "Haswell" to "Skylake" clock-for-clock.



Between "Skylake" and "Coffee Lake," Intel turned its R&D efforts toward refining the 14 nm process. It met with success on "Kaby Lake," and owing to its significantly higher clock speeds, "Kaby Lake" was able to provide higher performance than "Skylake." With "Coffee Lake," the nominal clock speeds look low, but Turbo Boost frequencies are higher than "Kaby Lake," and refinements in the process allow the chip to sustain elevated boost-clock states better. As we mentioned throughout the introduction, the design focus of these chips is to increase core counts across the board in order to better compete with AMD Ryzen.

The Gen 9.5 integrated graphics core takes up nearly a third of the die-area. Since it's of the same core-configuration as the one on the "Kaby Lake" silicon, it still features 24 execution units in the GT2 trim (featured on the i7-8700K). Higher clocks coupled with some driver-magic let Intel brand it "UHD Graphics." Don't expect to play PUBG at 4K on this, the "UHD" moniker only indicates that the IGP can handle 4K Ultra HD displays, features modern connectivity options such as DP 1.4 and HDMI 2.0, and can playback 4K video in new formats with 10-bpc color and HDR10/Dolby Vision standards.

The Z370 Platform

Intel is launching its first Core "Coffee Lake" processors alongside the Z370 Express chipset. Less expensive B360 Express and H310 Express models will be introduced in Q1-2018. The Z370 Express chipset, which succeeds the Z270 Express, appears to carry over the same platform feature set. It wouldn't surprise us if the Z370 turns out to be a re-brand of the Z270; however, we have no way of telling right now. Perhaps we will deal with this question in our upcoming motherboard reviews. As we mentioned on the previous page, 8th generation Core processors won't work on 100-series/200-series chipset motherboards.



This chipset compensates for the narrow 16-lane PCI-Express root-complex of the CPU by putting out 24 general-purpose PCI-Express gen 3.0 lanes. That's a good enough lane budget for up to three 32 Gbps NVMe devices (M.2 or U.2), the x4 and x1 PCIe slots, and even bandwidth-hungry onboard devices such as Thunderbolt 3 controllers, additional USB 3.1 gen 2.0 controllers, 10 GbE network controllers, etc. The integrated SATA controller puts out six SATA 6 Gbps ports with RAID support; the chipset also puts out two 10 Gbps USB 3.1 gen 2.0 ports and up to eight 5 Gbps USB 3.1 gen 1.0 ports, besides a number of USB 2.0 ports.