Introduction

Finally, here's our Intel Core i9-10900K 10-core processor review. Our samples, which were shipped by Intel over two weeks ago, took forever to reach us because DHL Air Freight was unable to find decent connections and had to reroute the shipment via several stops.

The Core i9-10900K is based on Intel's new "Comet Lake" microarchitecture that made its debut with the mobile 10th generation Core series. It is hopefully the last derivative of the company's 4-year old "Skylake" microarchitecture. When Intel launched its 6-core "Coffee Lake" silicon as its first response to "Zen," we had hoped that would be the last iteration of "Skylake," which wasn't to be. The following year, Intel increased the core count to eight, hoping it would be unbeatable by AMD. "Zen 2" surprised everyone, not just with its double-digit IPC gains, but also core count increases to 12 and 16 on the mainstream AM4 socket. Evidently, Intel's 10 nm nodes aren't ready for volume production of desktop processors, and so Intel went back to the drawing board to stretch the silicon one last time by adding two more cores. This 10-core die is hopefully Intel's last because we doubt adding more cores makes sense on this 14 nanometer process—as we'll see in this Core i9-10900K review, power draw is high enough already.



With Comet Lake, Intel introduces a new desktop platform built around the new LGA1200 socket and Intel 400-series chipset. Development of a new package isn't just consistent with Intel's roadmap of spanning a new desktop platform with no more than two processor generations, but also necessitated by the steep power requirements of the Core i9-10900K. To stand a chance against AMD's "Zen 2," it wasn't enough that Intel just add more cores, they also had to beef up the power delivery capability of these new processors to sustain higher clock speeds, which explains why more pins are required.

Besides just increasing the number of cores to ten on the Core i9-10900K flagship, Intel also added innovation with clock-speed management, using up to three separate boosting algorithms. This helps the Core i9-10900K hold on to higher clock speeds, and spread the increased frequencies among its cores better.

Generally, with Comet Lake, Intel is shoring up multi-threaded performance across the lineup. The Core i9 series is now 10-core/20-thread with the introduction of the new die. The Core i7 series, which was 8-core/8-thread with 12 MB L3 cache in the 9th generation, is now 8-core/16-thread with 16 MB L3 cache (identical configuration to the 9th generation Core i9-9900). The Core i5 series now consists of 6-core/12-thread parts with 12 MB L3 cache (identical configuration to the 8th generation Core i7-8700), up from its 9th and 8th generation predecessors being 6-core/6-thread parts with 9 MB L3 cache. The 10th generation Core i3 chips are 4-core/8-thread with 8 MB L3 cache (identical configuration to the 7th generation Core i7-7700), up from its predecessors being 4-core/4-thread. See what Intel did here? It tapped into previous-gen hardware IP it already had, re-positioning them at progressively lower price points.

In this review, we have with us the Core i9-10900K, the flagship "Comet Lake-S" part priced at $500. This 10-core/20-thread processor is capable of boosting up to 5.30 GHz, and Intel claims that it will be the "world's fastest gaming processor"—we'll check on that in this review. Its nominal clock speed is 3.70 GHz. Intel has also relaxed several design restrictions on power limits and boosting behavior, letting motherboard designers come up with basically their own boosting mechanisms, which are enabled by default on many boards we've seen. The out of the box performance will hence wildly vary from board to board unless you tell the BIOS you want it to run at Intel specs at first boot.

To bring you the complete picture of what performance to expect from the Core i9-10900K, were presenting three settings in this review:

• One that sticks to Intel specs (125 W power limit)
• One that maxes out the turbo headroom (effectively no power limit); frequencies are still controlled by Intel algorithms
• All-core manual overclock to 5.1 GHz fixed on all cores with all power limits disabled

A Closer Look

Our Core i9-10900K and Core i5-10600K review samples came in a fancy special packaging meant for reviewers only. The retail Core i9-10900K comes in an upscale-looking paperboard box with a large acrylic ornament to revel upon the chip inside.

Intel does not include a cooling solution in the retail box. You're expected to use any aftermarket cooler capable of handling thermal loads of up to 125 W TDP. As you'll find out later in the review, cooling has a direct impact on performance.


The Core i9-10900K looks like any LGA1xxx processor released by Intel in the past decade. The processor is only compatible with socket LGA1200 motherboards, as the round notches on the sides have been moved. It will not fit on any older motherboard.


Luckily, socket LGA1200 retains cooler compatibility with all older LGA115x-series sockets. This means you're going to be spoiled for choice picking a cooler to go with this processor.

Architecture

Under the hood of the Core i9-10900K is the 10-core "Comet Lake-S" silicon, built on the same 14 nm++ process as the previous two generations, but with one key difference.


Intel physically thinned the die along the Z-axis by removing as much of the silicon bulk as possible, the idea being that heat generated by the die travels through less silicon. The more conductive copper integrated heatspreader has been made thicker, and a solder thermal interface material (STIM) is used between the two. The die area is estimated by AnandTech to be around 198.4 mm², and from the looks of it, Intel can't add any more cores on this package by stretching the die without chopping off the iGPU.


The "Comet Lake-S" silicon is laid out similar to the past four generations of Intel mainstream processors, with two rows of CPU cores flanked by the iGPU on one side and the system agent (integrated northbridge) on the other, and a Ringbus Interconnect serving as the town square between the various components. The last-level cache is scattered across as slices of 1 MB or 2 MB, adding up to 20 MB of unified L3 cache that's equally accessible to all cores. It's fascinating to see how Intel retained its Ringbus-based multi-core design for this generation instead of designing something with its new Mesh interconnect. Intel still hasn't reached the core-count barrier beyond which the Ringbus has to be junked in favor of Mesh Interconnect tiles, lest it suffer the detrimental effects of ringbus latencies.

Much of the processor's uncore components are clumped into the System Agent, which contains the memory controller, PCI-Express gen 3.0 root-complex, DMI interface, and memory PHY. On the other end of the ringbus is the Gen 9.5 integrated graphics, which has practically been carried over for the past three generations, featuring 24 execution units in the GT2 trim. All SKUs in the desktop 10th gen processor series appear to have the top GT2 IGP. 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 core itself is identical in design to "Skylake," and hence there are no IPC increases to be had. As we explained in the introduction, all of Intel's efforts to increase gaming, single-threaded, and less-parallelized application performance revolves around increasing the clock speeds, and deploying as many as three intelligent boosting algorithms to achieve the advertised clock speeds.

Boost Algorithms

The Core i9-10900K has a nameplate base frequency (aka nominal clock) of 3.70 GHz. The processor embarks on a fascinating journey toward 5.30 GHz from here. The processor leverages Turbo Boost 2.0 to raise clock speeds up to 5.10 GHz, with varying boost multipliers along the way to workloads scaling across multiple cores. Beyond this, the Turbo Boost Max 3.0 algorithm, carried over from the company's Core X HEDT processors, is engaged. The algorithm raises frequencies of the processor's two favored cores up to 5.20 GHz. Beyond this, the processor invokes Thermal Velocity Boost, a brand-new algorithm that opportunistically overclocks the processor to 5.30 GHz in short bursts provided the processor's cooling solution is effective (able to keep temperatures below 70°C), and the processor's power budget permitting.

With "Comet Lake," Intel is introducing favored cores to its mainstream desktop processor segment, which was until now restricted to its HEDT and server processors. AMD supports this tech across its Zen product stack. Two of the best-performing cores on the silicon are identified at the time of each die's manufacturing, their information is hardcoded. Favored cores-aware operating systems (Windows 1709 or later and Linux kernels since January 2018) are able to identify this information, and send the most traffic to these two cores, as they are able to sustain boost frequencies best.

Intel has significantly raised power limits to choreograph the three boost algorithms. PL1, or power level one, spreads a 125 W power budget to support them across a default 56-second Tau (a time value for which the processor is allowed to sustain its boost frequencies before having to fall back to nominal clocks). PL2, on the other hand, is something else. Set at 250 W, Motherboard manufacturers have been given the freedom to override PL2 and Tau as they see fit, and so various motherboards implement power limits differently, depending on the effectiveness of their VRM solution. You are hence rewarded for buying a pricier motherboard that has a better VRM coupled with a more aggressive vendor-supplied power-management. Of course, all boards come with Intel-spec fallbacks. We will examine the boosting behavior at Intel spec and compare it with the "ASUS enhanced" specs of the ROG Maximus XII Extreme later on in this review.

Intel introduced a handful of overclocking enhancements with the 10th generation, including the ability to toggle HyperThreading on a per-core basis rather than globally. This could be an interesting option for those gaming and streaming, with a certain number of cores having HTT disabled for the best gaming performance, certain cores having them enabled, and Windows process core-affinity settings taking care of the rest.

The company also introduced the ability to overclock the DMI chipset bus. DMI is a PCIe-based interconnect that handles transfers between the processor and the chipset (PCH). The LGA1200 platform uses DMI 3.0 (comparable to PCI-Express 3.0 x4 in terms of bandwidth). Intel has apparently decoupled PCIe clock domains to enable you to overclock the DMI and PEG (that topmost x16 PCIe slot) without destabilizing your PCIe setup for graphics cards. The refreshed Extreme Tuning Utility now comes with finer-grained control of the voltage-frequency curve. The company also updated the Performance Maximizer app that automates overclocking using the trial-and-error method.

The Z490 Platform

Z490 is the top 400-series chipset targeted at gaming desktops and PC enthusiasts, as it enables serious overclocking and multi-GPU support. In terms of I/O capabilities, the chipset is nearly identical to the Z390, with 24 downstream PCIe gen 3.0 lanes, six SATA ports, six USB 3.2 gen 2 ports (that can be converted to three USB 3.2 gen 2x2 ports), ten USB 3.2 gen 1 ports, and fourteen USB 2.0 ports. Intel is recommending its i225-V 2.5 Gbps Ethernet chip as the wired networking solution to go with Z490, and the company's AX201 802.11ax WiFi 6 WLAN solution to go with the chipset's CNVio interface.