Intel i9-13900K and i5-13600K review: Beating AMD at its own game

When AMD brought the first Ryzen processors to desktops in 2017, the chips’ secret weapon was cores. AMD couldn’t match Intel’s number of instructions-per-clock or hit the same clock speeds, but what it could do was sell you six or eight reasonably performant processor cores at the same price Intel charged for four.

The 12-core and 16-core options came a couple years later, bringing high-end workstation performance to much less expensive PCs—and forcing Intel to follow suit. That brought down costs. In 2014, an 8-core CPU cost $999 on top of a pricey workstation motherboard; today, you can get 8-core chips for less than $300 and pop them into an $80 board if you want.

Five years after Ryzen debuted, the shoe is on the other foot. AMD now has the technological lead, in terms of instructions-per-clock, CPU manufacturing, and power efficiency. But by throwing more cores at the problem and pricing its chips aggressively, Intel has put together a desktop lineup that’s easier to recommend than AMD’s, even though AMD just fielded its fastest desktop processor lineup ever.

“More cores” isn’t all there is to Intel’s 13th-generation “Raptor Lake” CPUs, but it’s pretty dang close. Given the similarities to the previous chip generation and the increase in Intel’s topline power consumption figures, I didn’t come to these new CPUs expecting to be impressed. Color me wrong—Zen 4 and Ryzen 7000 are more interesting, but Intel’s 13th-gen chips are a lot easier to recommend.

P- and E-cores (re-)explained

When you hear “efficiency core” (E-core), you might think of a low-performance CPU core used to reduce power usage in laptops. That’s not necessarily incorrect, but what’s truly efficient about Intel’s E-cores is the amount of space in a CPU die that they take up. In Intel’s illustrations and in die shots of the Alder Lake and Raptor Lake chips, a cluster of four E-cores (and they always come in groups of four, since each group of four cores shares cache and other resources) takes up about the same amount of space as a single performance core (P-core).

The P-cores and E-cores help with different kinds of workloads. P-cores handle games and other apps where one or just a handful of fast cores are needed, while the E-cores collectively make a big difference when chewing through tasks that can engage all of a CPU’s cores at once.

That approach has drawbacks as implemented. Only Windows 11 and newer Linux kernels support Intel’s “Thread Director,” a software layer that helps decide which tasks are assigned to E-cores and which are assigned to P-cores; Windows 10 and older operating systems will run on these hybrid CPUs, but not always optimally. Other apps that request low-level hardware access (such as anti-cheat or DRM software, old games, restrictive test-taking software) can crash or fail to run on these hybrid CPUs, though many of these issues have been fixed since Alder Lake’s release.

And the new Intel CPUs don’t support AVX-512 instructions, something Intel had been pushing in its CPUs for years. AVX-512 is present in the P-cores but permanently turned off because the E-cores don’t support it. This means AMD’s newest CPUs support instructions that Intel’s don’t, even though Intel originated them and pushed their adoption for years (they are, admittedly, only beneficial in a handful of relatively niche workloads).

All things considered, though, for most workloads and most computers, the combination of P- and E-cores works, and the arrangement gives Intel a way to stay competitive with AMD’s single- and multi-core performance without needing to cram 12 or 16 cores into a single giant piece of silicon. Unlike AMD, Intel isn’t using chiplets in its consumer desktop CPUs.

New to Raptor Lake

This brings us to Raptor Lake. The E-cores are architecturally very similar to the ones included in 12th-generation processors, but they are running at somewhat higher clock speeds, each cluster of cores gets 4MB of L3 cache instead of 2MB, and (most importantly) there are a lot more of them. The Core i9-13900K goes from 8 cores to 16, while the i7-13700K and i5-13600K both go from 4 to 8.

To accommodate these extra cores, Intel boosted the power limits for all of these processors, upping the maximum Turbo Boost power limit (also called PL2) from 241 W to 253 W for the i9; from 190 W to 253 W for the i7; and from 150 W to 181 W for the i5.

Changes to the P-cores are similarly minimal, though there have been some small tweaks to help them keep pace with what AMD is doing. Clock speeds are higher across the board, with 200, 400, and 600 MHz boosts for the i5, i7, and i9, respectively. These increases were possible thanks in part to a higher power limit and a maturing Intel 7 manufacturing process, though Intel told us that optimizing the layout of the CPU die and “improving the speed paths” also helped.

The P-cores are aided by another L3 cache boost, from 1.25MB per core in 12th-gen Alder Lake CPUs to 2MB per core in Raptor Lake.

Intel officially supports faster DDR5-5600 RAM with Raptor Lake (while still maintaining the same support for DDR4-3200). Technically, using XMP to overclock your RAM could get you support for these and faster speeds with 12th-gen CPUs, making this a less impactful upgrade. Anecdotally, I’ll say that the DDR5-6000 kit that we used in our testbed won’t boot with 12th-gen CPUs unless we set it to run at DDR5-5800, while it works fine at its native speed with a 13th-gen CPU in the same motherboard. Your experience may be different with different RAM and different motherboards.

All of Intel’s 13th-generation Core i5, i7, and i9 chips should eventually benefit from additional E-cores and clock speed boosts, with the Core i3 tier being the last to include only P-cores. Rumors suggest that lower-end CPUs won’t be getting the same cache increases as the high-end parts, but we’ll see what Intel actually announces when it’s ready to take the wraps off the rest of the Raptor Lake lineup.

Our testbed

Our CPU testbed systems remain mostly the same as they were for our Ryzen 7000 review. We’ve compared the i9-13900K and i5-13600K to the previous-generation i9-12900K and i5-12600K, and to AMD’s flagship 16-core Ryzen 9 7950X and the entry-level Ryzen 5 7600X (with some numbers from the 7700X thrown in occasionally). Intel didn’t provide review samples for any i7 chips.

All our CPUs are tested in a Lian Li O11 Air Mini case with an EVGA-provided Supernova 850 P6 power supply and a 280 mm Corsair iCue H115i Elite Capellix AIO cooler.

We use an AMD-provided Radeon RX 6800 as our GPU, which means many benchmarks are still GPU bottlenecked before they become CPU bottlenecked. If you’re building a gaming system on a budget, keep in mind that a modern Core i5 or Ryzen 5 CPU isn’t going to hold a midrange GPU back—the CPU subscores may differ, and minimum framerates might improve with a better CPU, but note that for Shadow of the Tomb Raider and Forza 5, literally all of these chips averaged the exact same framerates (between 186 fps and 188 fps for Tomb Raider, and 126 fps or 127 fps for Forza). We’re hoping to change this GPU out for a more powerful one in the near future.

You might also notice that we used a last-generation Z690 motherboard while testing the 13th-generation CPUs; all motherboards based on a 600-series chipset should be capable of supporting 13th-gen CPUs as long as your board manufacturer releases a BIOS update.

Intel told us that an updated Z690 board should perform identically to a similarly configured Z790 board, and our experience bears that out. Z790 adds more I/O options, but there’s nothing inherently better about its performance. Cheaper boards based on H610 and B660 chipsets might limit performance more, but that has more to do with the motherboards’ cost than their chipsets; cheaper boards will have less robust power delivery systems and smaller voltage regulator module (VRM) heatsinks that could keep the high-end CPUs from getting as much power as they can use.

Finally, modern desktop CPUs in most non-OEM motherboards have user-configurable power limits, which can reduce the amount of power the CPU is allowed to consume. Benefits include lower heat generation and power usage. The drawback is reduced performance, though it’s not generally proportionate (a chip using twice as much power may only run 20 or 30 percent faster, when all the cores are busy at once). Multi-core performance is much more impacted by lower power limits than single-core performance, so lowering the power limits is especially worth considering for people primarily using their PCs to play games. We talk more about Intel’s power settings in our Core i7-12700 review, and more about AMD’s power settings in our initial Ryzen 7000 review.

Many, many cores make light work

Comparing 12th-gen chips to 13th-gen ones, Intel says the i9-13900K should be around 15 percent faster than the i9-12900K in single-threaded tasks and 40 percent faster at multi-threaded workloads. Our testing largely bears this out, though the effect was more pronounced in the synthetic benchmarks we ran than it was in our gaming and video encoding tests.

Performance gains at the i5 tier are a bit less significant—single-core performance gains are in the low single digits, and multi-core improvements fall within the 25 to 40 percent range, depending on the benchmark. If you have a 12th-gen system and are considering an upgrade, it makes more sense if you’re also jumping up a tier or two (from an i5 to an i7 or i9).

These year-over-year performance gains are generally enough to put Intel back ahead of AMD in many of our tests, though again, it depends on which CPU tier you’re talking about and which specific test.

These chips’ single-core performance reflects a difference in approach between AMD and Intel. Intel’s lower-end CPUs have lower single- and multi-core performance, so even if you only care about single-core performance, you need to spend money on a higher core-count processor to get it. The Ryzen 7000 chips’ single-core performance is pretty similar up and down the entire product range, an approach Apple has taken with its M1 chips.

The upshot is that Ryzen 7600X is level with or slightly above the i5-13600K’s single-core performance in our tests, while the i9-13900K manages to beat the Ryzen 9 7950X by around 10 percent. The differences in our game benchmarks don’t tell a consistent story—Borderlands 3 and Forza 5 seem to favor AMD’s chips regardless of raw speeds, while Tomb Raider and Final Fantasy XIV do the same for Intel. The main takeaway should still be: Don’t overpay for a CPU that’s just going to play games, unless you’re going to use it as some kind of photo/video editing or CPU rendering workstation. These CPUs will all feel essentially the same in most games with most graphics cards.

Our CPU-based video encoding test engages all of these processors’ cores for a more extended period of time, and it’s here where Raptor Lake can pull ahead decisively, especially at the Core i5 tier. The i5-13600K encodes our H.264 sample video in 71 percent as much time as it takes the Ryzen 5 7600X, and it takes 85 percent of the time to encode the H.265 sample video. The Zen 4 architecture seems to handle H.265 video better than Intel’s cores overall, but Intel will sell you six big cores and eight little ones for the same money that AMD will sell you six big cores all by themselves. Those extra computing resources are hard to beat. That’s also visible in the Geekbench and Cinebench scores.

The competition is closer for the top-tier CPUs, where the Ryzen 9 7950X’s 16 big cores are better able to keep pace with the Core i9-13900K’s combination of eight large cores and 16 small ones. The i9-13900K is generally faster in our benchmarks, but by around 10 percent rather than 45 to 60 percent. And the 7950X makes quicker work of the H.265 sample video than the i9-13900K.

Power use: High by default

The Core i9 and Ryzen 9 chips aren’t too far apart in performance, which is worth keeping in mind when you see how much power the Core i9 needs to do its job. Let’s set TDP numbers aside and compare actual average CPU package power consumption during our H.264 video encoding test:

The Core i9-13900K can encode our sample video in 87 percent of the time it takes the Ryzen 9 7950X to do so, but it uses 129 percent of the power to get the job done—it finishes faster, but not so much faster that it can make up for the fact that it’s drawing 251 W of power while the Ryzen chip draws 169 W.

You can curtail the Intel chips’ power consumption—we tested the i9-13900K at 125 W (the chip’s base power level), and we tested both Intel chips at 65 W (Intel’s default base power level for non-K desktop CPUs). Halving the i9-13900K’s power consumption didn’t come anywhere close to halving its performance. Cinebench is fairly representative. The default power settings offer roughly 30 percent better performance for 200 percent of the power. We also noticed much less of a difference in our single-threaded and gaming tests, since these tasks don’t hit power limits as often as multi-threaded tests do.

But AMD’s manufacturing advantage really comes in handy here. The 7950X set to a 65 W TDP level (consuming about 90 W of power, since AMD’s TDP figures are more abstract than Intel’s base power figures) runs as fast or faster than an i9-13900K consuming 125 W of power in our H.264 and H.265 encoding tests, respectively.

While it’s frustrating that the incentive for both companies right now is to drive up power consumption in an effort to one-up each other and to keep posting year-over-year performance gains, the non-K, non-X processors that end up in the vast majority of PCs should all have more efficiency-minded default settings. Testing these CPUs at lower power levels shows that (1) they can all be reasonably power-efficient under the right circumstances, though AMD has the advantage, and (2) that Intel continues to struggle a little with power consumption because of ongoing post-tick-tock-era manufacturing challenges. (Intel does say that it has a plan for reclaiming its leadership in chip manufacturing, to which I say: Great! Prove it.)

Turning the tables on AMD

For what is in many ways a minor refresh, Intel has achieved something impressive with Raptor Lake: keeping pace with and even beating AMD’s latest Ryzen processors’ performance with only modest architectural changes and no manufacturing process improvements. As these manufacturing improvements get more and more difficult to pull off, chipmakers will need to get more creative to keep boosting performance, and the hybrid architecture approach is effective here.

It certainly helps Intel that it can be several hundred dollars cheaper to put together a Raptor Lake system than an AM5-based system, thanks to inexpensive and plentiful 600-series motherboard options and continued support for DDR4 memory. As if to emphasize this, the midrange Core i5-13600K absolutely thumps the Ryzen 5 7600X’s multi-core performance for the same or less money. The i9-13900K isn’t quite as big an improvement over the Ryzen 9 7950X, relatively speaking, but it still matches or beats AMD’s flagship in most of our tests for around $100 less.

The good-ish news for AMD is that its singe-core performance is better than Intel’s on the low end, and the i9-13900K especially needs to use gobs more power to beat the 7950X decisively. That might not matter for people who care about performance more than they care about anything else, but if you’re constantly pushing your processor, the power savings can add up over time.

The bad news for AMD is that to compete on price, the company’s easiest move in the short-to-medium-term is probably to push the prices on its midrange chips back down to around where they were three years ago—six cores for $200 and eight cores for $330. To compete on performance, it may need to start fielding some kind of hybrid architecture of its own (sketchy rumors currently suggest that a chip with Zen 5 P-cores and cut-down Zen 4 E-cores is a possibility).