The Core i9-10900K is the tip of Intel's spear in its newest 10th Generation Core ("Comet Lake-S") lineup of desktop CPUs.
As a top-of-the-line 10-core processor (the estimated selling price is $488 for the bare chip), the Core i9-10900K posts great scores in gaming with our standard test GPU (even setting a record one title, Counter Strike: Global Offensive).
But it doesn't quite match up with AMD's competing 12-core Ryzen 3900X in content-creation and highly multi-threaded tasks.
The need for a new motherboard to upgrade to Comet Lake-S only adds to the real-world cost of the Core i9-10900K, and its thermal demands left us wanting.
If you already have a late-model gaming or content creation PC, it doesn't do enough, at least at the moment, to justify the cost of adopting a whole new platform, especially in a world with strong Ryzen 9, Ryzen 7, and Ryzen 5 chips in AMD's arsenal (plus the spanking-new Editors Choice Ryzen 3 3300X for gaming budget buyers).
The peak Comet Lake-S CPU doesn't change the overall desktop-CPU calculus much.
If you're in the market for a mega-core monster CPU for gaming and creative work on a mainstream platform, and don't want to jump to the Intel Core X-Series or AMD's Ryzen Threadrippers, the recommendation is the same today as it's been since July of last year: Go with one of AMD's mainstream AM4 chips instead.
Comet Lake-S: Reheating 14nm
The Core i9-10900K is, like all other Comet Lake-S chips in the full line, based on the same 14nm++ production process as 9th Generation desktop chips were.
(See our overview of the family's reveal earlier this month.) It comes at an interesting time for Intel, representing what could be (read: hopefully) the end of an era for its 14nm lithography.
The whole Comet Lake-S stack makes use (now much refined, of course) of the same basic process technology as "Broadwell" back in 2014.
With AMD's Zen 2 CPUs, introduced from the second half of 2019 and into 2020, all integrating a 7nm chiplet-based architecture, it's clear that Intel needed to pull out all the stops for 10th Generation if it hoped to remain relevant.
After all, AMD is reportedly preparing for the launch of its Zen 3 architecture later this year.
The new Intel Core i9-10900K represents the top of an entirely new stack for Intel, reaching all the way down low-power-efficiency CPUs like the Core i3-10100T.
The company has reintroduced its thread-doubling Hyper-Threading technology to many of the chip SKUs, a fan favorite that was broadly disabled in most 9th Generation desktop chips.
With the launch of Intel's new 10th Generation stack of chips has come a new socket, LGA 1200, and a gigantic pile of new of motherboards to support it based on the new Z490 chipset.
(Boards based on the cheaper B460 and H410 chipsets will follow in short order.)
Whether or not these 14nm++-based chips had be upgraded to all-new chipsets and series of platforms is Intel's own calculation, but it's clear that the company is doing all it can to squeeze as much juice out of 14nm before it's retired sometime in, perhaps, 2021.
Given the chip maker's much discussed issues with 10nm production, this new platform may be the final stretch of necessary 14nm highway before the bridge to a much smaller process technology.
The thing is, though, even with the new socket, new chipsets, and new platform, the whole works still feels behind AMD's latest, in some respects.
For one thing, Z490 doesn't bring any massive new must-buy features.
A second big reason: the lack of support for PCI Express 4.0.
Despite the fact that AMD has supported PCIe 4.0 since the launch of the X570 chipset back in July of 2019, in near-June of 2020 Intel still has no "official" plans to support the new standard on Z490 motherboards.
(We asked; it was a "no comment" situation.)
Now we say "official" in quotes because leaks correspond to some other indications from motherboard makers.
According to a leaked internal slide acquired by VideoCardz.com (take as ever, leaks with a grain of salt; they can be faked), Intel's upcoming plans for its next generation of desktop chips, dubbed "Rocket Lake-S," include PCIe 4.0 support on the Z490 platform.
Some makers of Z490 motherboards (including Asrock, Gigabyte, and MSI) have dubbed some of their Z490 motherboards PCI Express 4.0 "ready," though what exactly they mean by that remains to be seen. As a result, it's looking likely that Z490 isn't just here for 10th Generation Core desktop, which makes sense when you consider Intel's standard pace of socket upgrades over the years (that is, they last for two generations).
AMD, on the other hand, has been supporting AM4 for years now, and even just as we wrote this announced on Reddit its official plans to continue support for its coming Zen 3 CPUs on some of the later AM4 chipsets.
The company is working wonders to ensure that the most possible customers have an affordable way to use their processors on existing hardware, from the bottom of the stack to the very top (excluding Ryzen Threadripper, of course, which is its own animal and saw a socket and chipset switcheroo with its latest generation).
Point being: If you're planning to upgrade to the Intel 10th Generation chips as of today, at bare minimum you'll need to factor in the cost of a new motherboard, which for Z490, at the lowest end, may cost you $199 and up for options like the budget-focused Asus TUF Gaming Z490-Plus.
More-affordable boards that utilize the LGA 1200-compatible H410 and B460 chipsets are due to drop next week; however, users who go that route will have to sacrifice the option for overclocking, which precludes opting for the Core i9-10900K as a viable option.
The "K" as ever means it's unlocked for overclocking, and that is part of the appeal for some of the folks who buy at the top end of any Intel chip stack.
Intel vs.
AMD Comparison: Core i9-10900K vs.
Ryzen 9 3900X
Speaking of bucks, no Intel chip-stack launch would be complete without our value comparison to AMD.
Let's take a look at it versus its closest competition...
Hold that $499 price in thought for a moment on the Ryzen 9.
The Intel Core i9-10900K is a 10-core/20-thread CPU with an Intel UHD Graphics 630 integrated graphics processor (IGP) with a TDP of 125 watts, clocked to 3.7GHz at base.
Intel claims the chip is capable of hitting 5.3GHz on both single-core and all-core tasks as long as the thermal limits of its "Thermal Velocity Boost" technology aren't breached (more on that in a minute).
On the AMD side, the best point of comparison is the 12-core/24-thread Ryzen 9 3900X, at least on specs.
On current street prices though, an interesting dynamic was playing out as we wrote this.
AMD lowered the 3900X's official list price to $449, and several retailers, among them Amazon and B&H, dropped this previously-$499 MSRP processor all the way down to $409, which we imagine was done in response (at least temporarily) to Intel's pending launch and/or Memorial Day sales.
At that price, if it sticks (there is no indication that it will), it's difficult to even keep these two chips in the same category, a task that only gets more difficult once the benchmark numbers start rolling in.
Even the "full" $449 list price is very competitive.
Now, back to that boost.
According to the company line: "Intel Thermal Velocity Boost (Intel TVB) is a feature that opportunistically and automatically increases clock frequency above single-core and multi-core Intel Turbo Boost Technology frequencies based on how much the processor is operating below its maximum temperature and whether turbo power budget is available."
We weren't able to get a full answer on how this tech differs from what it sounds like it's describing (that is, automatic overclocking, which you can find on many higher-end motherboards these days), nor were we able to track any instances where this feature was specifically the reason that a clock was hitting a higher clock speed.
But it's all baked into our stock testing numbers, which we'll get into in a moment.
We tested the Core i9-10900K on an Asus ROG Maximus XII Hero Wi-Fi motherboard, with 16GB of G.Skill memory clocked to 3,000MHz (for comparability with our earlier CPU reviews), an Intel SATA boot SSD and an ADATA SATA M.2 secondary drive.
All this was packed in ADATA's XPG Invader chassis, fitted with a Deepcool GamerStorm Captain 240 EX 240mm liquid cooler and a 750-watt Corsair power supply. For our gaming tests, we used an Nvidia GeForce RTX 2080 Ti, at Founders Edition clocks, as we have on all recent mainstream and high-end CPU reviews.
Unlike most of Intel's mainstream CPU line, the Core i9-10900K does not bundle a cooler in the box, and though you technically could run it on a large, high-end aircooler like a Noctua model, we recommend liquid cooling for this chip and its 125-watt TDP that surges well over 200 watts in some high-load situations.
A 240mm all-in-one-style liquid cooler will set you back at least $100 or so unless you own one already; LGA 1200 works with coolers for recent Intel mainstream chip sockets, such as LGA 1151 and LGA 1156.
That is in marked contrast to the Ryzen 9 3900X, which does bundle AMD's nice Wraith Prism air cooler right in the box, further ka-ching savings.
(Only the step-up Ryzen 9 3950X comes cooler-less and needs a liquid cooling solution.)
Benchmark Rundown: It's a 10-Core War
First, some performance preface. We noted some more "teething" with this prelaunch testing session than with most new-platform-launch CPU reviews.
We suspect most of this wobble will clear with early BIOS updates; we can't help but think with the COVID-19 crisis and all of the disruption it has caused to commerce and business, that a smooth-as-glass launch could hardly be immune to at least a few hiccups.
We started our testing with a Gigabyte Z490 Aorus Master board provided by Intel with the chip samples, but we had difficulties getting the board to install Windows 10 on a couple of SATA SSDs; the install would fail, with the PC shutting down and rebooting partway through the Windows 10 setup process for reasons we were unable to diagnose.
We ended up settling on the Asus ROG Maximus XII Hero board as our alternate tester.
We had an initial issue with a Samsung 2.5-inch SATA SSD we installed as our boot drive there; that only resolved when we swapped it out for an Intel SATA 2.5-inch drive as our boot drive.
(Through it all we used a secondary ADATA SATA M.2 drive to host all our game install files.) Also, the Asus board received a couple of BIOS updates in the course of our testing, one of them (version 0509) coming so close to the launch that we were able to install it just two days before the launch date and rerun our tests, invalidating all our earlier results.
(It cleared up some thermal oddments we noted in early tests.)
We are still not 100 percent convinced that the board is pushing the i9-10900K to its very tip-top potential at stock settings, but we also used the board for some "AI assisted" overclocking that perked things up a bit further.
More on that later.
First, some stock-settings results.
Cinebench R15 and R20
Among the most widely used predictors of a CPU's relative performance are the Cinebench R15 and R20 benchmark tests, which offer a good muscle measure for demanding multi-threaded content-creation apps.
These are thoroughly CPU-centric tests that gauge both the single-core performance and the multicore performance of a processor when it is stressed.
The resulting scores are proprietary numbers that represent the CPU's capabilities while rendering a complex 3D image.
The all-cores Cinebench numbers, we should note, on this processor were a tad inconsistent, and we've reported the best runs we got at stock.
(The single-core numbers were rock steady.) On repeat runs we would often see the all-cores score drop by 3 to 5 percent, for reasons we were unable to nail down before the launch.
Still, you can see the edge held by the rival Ryzen 9 3900X here on the all-cores tests.
iTunes
For another kind of real-world look at single-core performance, we use a, shall we say, well-aged version of Apple's iTunes to encode a series of music tracks.
It remains in our test lineup simply as a representative of legacy software we all use from time to time that has not been heavily optimized for multicore operation.
Given the chance at a single-threaded task, Intel's chips are typically clear winners over equivalent AMDs, thanks to their higher single-core speeds.
That trend held fast here.
Then again, we hope you're not buying a $500 CPU and new motherboard for this kind of thing alone.
POV-Ray
The POV-Ray benchmark is a synthetic, highly threaded rendering test that offers a second opinion on the Cinebench results.
This test uses ray tracing to render (offscreen) a three-dimensional image.
(Note that it doesn't use the ray tracing features of Nvidia's RTX-class GPUs; this is purely CPU-focused.) We run all-cores and single-core variants.
The testing remained rock-steady across all runs of all-cores POV-Ray, unlike on Cinebench.
The story of multi-threaded dominance, though, continues for the less-expensive Ryzen 9 3900X here, though the chip is quickly shown up by the Core i9-10900K once the single-core POV-Ray run kicks off.
Handbrake and Blender
As an all-core rendering benchmark, the Handbrake test is a great indicator of how well a processor will handle tasks like video editing, video rendering, and video conversion, as these kinds of apps tend to munch on all the cores and threads they can get in their teeth...
Here the Ryzen 9 3900X, again, wins by a consequential amount.
If you're only converting the odd video file here and there, the roughly half-minute lead won't make much of a visible difference.
However, if you work in a field like digital imaging tech (DIT) and need to process terabytes of videos a day, those lost minute fragments will really start to add up.
Meanwhile, the shorter Blender test, as run with our test file, is mostly useful for highlighting the vast differences between low-end and high-end chips, and the similarities between chips within these two categories.
A win for the Intel Core i9-10900K in a multi-threaded task! Now, to be fair it's within the margin of error, but our Blender render is also considered a lightly-threaded task, which Intel does seem to do well in compared to AMD's high-core monsters.
7-Zip
And here on the 7-Zip file-compression benchmark, another thread-happy,...
