Intel Core Ultra 9 285K Processor Review - Gamers Will Want More

Introduction, Specifications and Pricing

The anticipation surrounding Intel's Arrow Lake-S desktop CPU family has steadily risen. And now it's here! We lead our coverage with the flagship Core Ultra 9 285K. It represents a massive architectural paradigm shift for Intel. Gone is the monolithic approach of previous Intel architectures in favor of a disaggregated chiplet based design. It's not the first such design from Intel. Meteor Lake and Xeon server chips have been built with tile designs, but it's a first for Intel on the desktop.

The Core Ultra 200S series, as it is formally known, replaces the 14th Gen Raptor Lake Refresh family. The new terminology should be easy enough to follow as long as Intel sticks with it for as long as it did from the 2nd to 14th Generation.

27

The Core Ultra 9 285K is an 8P-16E core design, the same as that of the high-end Raptor Lake Core i9-13900K and i9-14900K, but that's just about all these chips have in common. The 285K includes eight Lion Cove P-cores and sixteen Skymont E-cores. In particular, The Skymont cores receive a 32% IPC boost, meaning the 285K is poised to outperform the 14900K in heavily multi-threaded apps despite the P-cores dropping support for SMT (Simultaneous Multi-Threading). The P-cores receive a claimed 9% IPC uplift.

Intel has heavily focused on power efficiency and performance per watt for the Arrow Lake-S generation. The 285K's P-cores have a 3.7GHz base clock and a 5.7GHz boost clock, so it doesn't need extreme power to reach that 6GHz level of the 14900K. The E-cores have a 3.2GHz base clock and a 4.6GHz boost clock. Intel rates the 285K with a PL2 of 250W, just 3W less than the 253W of the 14900K, but as we'll see in testing, the reality is the 285K is indeed a more power-efficient chip. That means cooler temperatures and less of a need for a high-end cooler.

The P-cores include their own dedicated 3MB of L2 cache, while each of the four E-core clusters shares a 4MB L2 cache. The P-cores and E-core clusters share a common L3 cache, which totals 36MB. The total L2 and L3 cache adds up to 76MB. Alongside the general-purpose cores, Intel is bringing a Neural Processing Unit (NPU) to the desktop for the first time. Integrated graphics performance has dramatically improved. It is built on the Xe-LPG architecture and includes high-spec media and display engines, as well as DirectX 12 Ultimate support. The 285K's total die area is 243 mm², and its total transistor count is 17.8 billion.

The Intel Core Ultra 9 285 K's MSRP is $589, the same as the 14900K's launch price.

Further Intel CPUs Reading – Our Latest Content

• Intel Core Ultra 9 285HX vs. 285K CPUs Head-to-Head - Desktop Power in a Laptop Form Factor
• Intel Core Ultra 5 245K Processor Review - Team Blue Has Seen Better Days
• Intel Core i9-14900KS CPU Review

Best Deals: Intel Core Ultra 9 285K

	Today	7 days ago	30 days ago
	$579.98 USD	$519.99 USD	$519.99 USD	Buy
	$576.99 USD	$579.98 USD	-	Buy
	$800.14 CAD	$788.25 CAD	$716.13 CAD	Buy
	$809 CAD	$809 CAD	-	Buy
	£479	£479	£479.50	Buy
	$579.98 USD	$519.99 USD	$519.99 USD	Buy
* Prices last scanned 1/13/2026 at 9:16 pm CST - prices may be inaccurate. As an Amazon Associate, we earn from qualifying purchases. We earn affiliate commission from any Newegg or PCCG sales.

The Arrow Lake Architecture

Intel switches to chiplets

The key design feature of Arrow Lake-S is the shift to a disaggregated tile-based architecture. Alongside the yield advantages of producing smaller chips, this approach allowed Intel to mix and match process technologies for different core functions. It gives Intel modular flexibility with its choice of components. For example, in the future, it could release a CPU with a faster iGPU without the expense or complexity of redesigning an entire monolithic chip.

27

Arrow Lake-S features four primary tiles: the Compute tile, SoC tile, Graphics tile, and I/O tile. Additionally, there is a Foveros base tile that serves as an interposer, linking the tiles to each other and the base package and socket pads. This approach differs significantly from AMD's, whose core complex dies and I/O die are connected via the Infinity Fabric and package substrate. Intel says its approach provides better latency and reduces power.

Compute Tile

The most important tile is the Compute tile. It contains the P and E-cores, and cache.

27

The Compute tile is built on the advanced (and expensive) TSMC N3B process. That alone is a very interesting choice, as Intel surely would have preferred to use its own foundries to manufacture it. This tile includes eight Lion Cove P-cores and sixteen Skymont E-cores. As you can see in the block diagram above, The E-cores are arranged in four clusters of four cores. They're all connected via a ring-bus interconnect, which gives all cores access to the 36 MB of L3 cache.

This arrangement differs from Alder Lake and Raptor Lake, which grouped the P and E-cores entirely separately. This arrangement reduces the hot spot generated by the P-cores, and heat is better distributed across the tile. It's also designed to minimize inter-core latency.

SoC Tile

The SoC tile is manufactured with the TSMC N6 node. It's not a cutting-edge node anymore, as apart from the NPU, it doesn't include compute-heavy units. This tile includes the DDR5 memory controller and PCI-Express functionality. The memory controller supports up to 192GB of DDR5 memory, with native support for DDR5-6400. It supports CUDIMMS, paving the way to DDR5-10000 and beyond.

The SoC tile adds support for 20 PCIe 5.0 lanes. Users can now run a PCIe x16 graphics card and x4 SSD without compromise. Older Intel platforms supported Gen 5 SSDs, too, but it required taking 8 lanes away from the x16 slot. A further eight PCIe 4.0 lanes are used to connect to the chipset.

27

The NPU is inherited from Meteor Lake, making it the first mainstream desktop platform to include a dedicated NPU. It offers up to 13 TOPS of performance, which is well under the 40 TOPS requirement to meet Microsoft Copilot+ certification. However, Intel argues that most desktops will include a discrete graphics card, putting the system well over that 40 TOPS level.

AI is all the buzz these days, and though the software ecosystem is developing rapidly, it's not all-pervasive yet. Arrow Lake's NPU may come into its own in the coming months and years.

GPU Tile

The GPU tile contains the Xe-LPG integrated graphics functionality, which is a big step forward from Raptor Lake's UHD 770 integrated graphics. This tile is built on the TSMC N5P node. Isn't it interesting how Intel's tiles are built on different processes?

The 285K's graphics tile includes four Xe cores, giving it a total of 64 execution units. We ran some tests and were surprised by the performance uplift. We'll come to that shortly.

The I/O tile contains platform-oriented functionality, including Thunderbolt 4 and other connectivity support. It includes the media engine functionality, an impressive part of Arrow Lake's spec. It supports decoding up to 8K @ 60 Hz with 10-bit HDR and encoding as high as 8K at 120 Hz.

27

The Display Engine supports HDMI 2.1, DisplayPort 2.1, and eDP 1.4. Depending on the connectivity of the partnering motherboard, the IGP supports displays up to 8K at 60 Hz with HDR, up to four 4K displays at 60 Hz with HDR or four 1080p displays up to 360 Hz. That's not bad at all for integrated graphics!

The Core Ultra 9 285K and Test System

Core Ultra 9 285K

27

Our test CPU arrived in non-retail packaging. The picture above, provided by Intel, shows the packaging you can expect when you buy a Core Ultra 9 285K.

27

The CPU itself looks nearly identical to 12th, 13th, and 14th-generation CPUs. Even though Arrow Lake is not compatible with older motherboards (the notches are different, so they won't physically fit), the new CPUs are compatible with LGA 1700 coolers.

27

The backside offers 1851 pads. Compared to LGA 1700 chips, the pad area of LGA 1851 CPUs takes some of the area formerly occupied by the area in the middle of the CPU.

Test System

Sadly, our review CPUs were stuck in customs for nearly a week, severely limiting our testing time. To avoid rushing our testing, we made the tough decision to delay our review from the original embargo time and date.

We did not have time to do any overclocking testing or test with CUDIMM memory. We will release a separate article on these topics at a later date. Hopefully, some BIOS and software updates will arrive to iron out some performance oddities encountered during our benchmarking.

• Motherboard: MSI MEG Z890 Unify-X
• GPU: MSI GeForce RTX 4070 Ti Super Ventus 3X - Buy from Amazon
• RAM: G.Skill Trident Z Neo RGB 2x16GB DDR5-6000 CL30 - Buy from Amazon
• Cooler: Cooler Master MasterLiquid PL360 Flux - Buy from Amazon
• OS Storage: Teamgroup T-Force Z540 2TB - Buy from Amazon
• Power Supply: Super Flower Leadex Titanium 850W - Buy from Amazon
• OS: Microsoft Windows 11 Pro - Buy from Amazon

Benchmarks - Rendering and Encoding

Cinebench 2024

Cinebench is a long-standing render benchmark that has been heavily relied upon by both Intel and AMD to showcase their newest platforms during unveilings. The benchmark has two tests: a single-core workload that utilizes one thread or 1T and a multi-threaded test that uses all threads or nT of a tested CPU.

27

Kicking off our testing of the 285K, we see an excellent result. Its nT test result of 2,437 puts it ahead of every other desktop CPU, and its single thread result of 141 narrowly puts it on top, too.

Blender

A rendering application like Blender is just one of many reasons a user will consider a high-core-count CPU like a Core Ultra 9 285K. We use the Whitelands demo file and record how long it takes to render the image.

27

It's another win for the 285K. It completed this test over 30 seconds quicker than the 9950X runner-up. Time is money, so the saying goes.

Handbrake

Handbrake is a simple-to-use video encoding and transcoding application. Here, we convert a 4K movie trailer to 1080p. The results below show the average FPS, where a higher result means the task will take less time to complete.

27

The 285K wins another battle here. Handbrake, Cinebench, and Blender all demonstrate the 285K's excellent performance under heavily threaded workloads.

Benchmarks - File Compression and Memory Latency

7Zip

7Zip is a commonly used free file compression and decompression app. It's susceptible to memory speed and latency changes and scales with the number of CPU threads.

27

Here, we see the first of several odd performance results. The 285K fails to match the 14900K. Perhaps the lack of hyperthreading does few favors in this test.

AIDA64 Memory Latency

Memory latency is an area that's traditionally favored Intel and its monolithic designs. AMD's chiplet architecture and the Infinity Fabric link inevitably add some latency compared to in comparison. A nanosecond or two here or there is not noticeable, but more significant margins, particularly when the memory is frequently accessed, will result in more undesirable and cumulative idle cycles.

27

Low memory latency is important for consistently high gaming performance. With DDR5-6000 C30 memory, both Arrow Lake chips fall well behind. Is this a result of placing the memory controller on a separate tile from the core complex? Or is it something that can be addressed with Windows or firmware updates? We'll have to wait and see.

Benchmarks - PCMark and 3DMark

PCMark 10 Productivity

We'd love to use our PCs purely for leisure, but some of us have to work, too! The PCMark 10 productivity test performs a series of tests using office productivity applications.

27

The Arrow Lake chips struggle in this test. Some of the subtests are more memory agnostic than others, so raw single-threaded performance is not the only determining factor in the result.

3DMark Time Spy Extreme

Time Spy Extreme is losing favor as a graphics benchmark in favor of Speed Way and Steel Nomad, but its CPU test is still a good measure of multi-core performance.

27

The Time Spy Extreme CPU test again shows the Core Ultra 9 285K's multi-threading prowess. Though not shown here, its graphics score was lower than expected.

Benchmarks - Gaming

Cyberpunk 2077

Cyberpunk 2077 is brutal on graphics cards, but when things like ray tracing are removed, it becomes more sensitive to CPU and memory performance differences.

27

Here, our RTX 4070 Ti Super is still quite GPU bottlenecked. It serves as an example of how virtually all modern CPUs are sufficient for gaming in such scenarios. Nevertheless, the 285K returns one of the lower results, indicating not all is well with its gaming performance.

Far Cry 6

Far Cry 6 is an example of a game that exhibits CPU bottlenecking with powerful graphics cards. It tends to perform well on Intel CPUs versus their AMD counterparts.

27

This is merely an OK result for the 285K. It still trails the 7800X3D and falls well short of the two 14th-generation CPUs.

Horizon Zero Dawn

Horizon Zero Dawn can achieve high frame rates with powerful graphics cards when using the' favor performance' preset.

27

Uh-oh. This isn't a good result at all. The 285K can't match the i5-14600K here, and the AMD chips - particularly the 7800X3D - are well in front.

Metro Exodus Enhanced Edition

Metro Exodus received an update that improved DLSS support, enhanced ray tracing features, and variable rate shading, among other things. Still, with a powerful graphics card, it is affected by CPU and memory performance at 1080p, though less so with a card like the RTX 4070 Ti Super.

27

Oh, Intel. What's going on here? Arrow Lake is not performing well in gaming right now.

Tiny Tina's Wonderlands

I'm a big fan of the Borderlands series and really enjoyed playing Tiny Tina's Wonderlands. At 1080p with the high preset, our RTX 4070 Ti Super still somewhat limits it, so the differences here are not as large as you'd see with an RTX 4090.

27

The difference from top to bottom here isn't as dramatic as some of the other games we tested, but the 285K still fails to impress.

Benchmarks - Integrated Graphics

F1 22

F1 22 is not a particularly demanding title, and it doesn't need a high-end graphics card to enjoy smooth frame rates. We tested at 1080p with high settings.

27

This result shows how much the 285K's integrated graphics capabilities have improved. Many consider 36 FPS to be playable. For laughs, we ran the 245K with AMD's FSR enabled, giving it a small but tangible performance boost.

Far Cry 6

As Far Cry 6 is not overly demanding on GPUs, it has been proven to run well on AMD's highly capable 8000G APUs.

27

Far Cry 6 is borderline playable at high settings on the 285K, but if you add some upscaling to the mix, it'll cross well over the 30FPS threshold. It'll even beat out the mighty 8700G if you do that.

Tiny Tina's Wonderlands

Tiny Tina's Wonderlands still asks a lot of integrated graphics. Even the 8700G is barely playable.

27

It's a step too far for the 285K here. Of course, few people will buy the 285K for its integrated gaming performance, but with the right settings and a dose of upscaling, Arrow Lake CPUs can be used for casual gaming, particularly if you're into playing popular online titles or MOBA games.

These results show Arrow Lake's Xe graphics have improved by 100% or more compared to the UHD 770 graphics of 14th Generation processors. It's hard to complain about 100% generation-on-generation performance gains!

Benchmarks - Power Consumption and Temperatures

Power

Intel is heavily touting the improved power efficiency of Arrow Lake, so how does it do?

27

A peak reading of 225W is a solid improvement over the 14900K. Even though it's not entirely representative of the 285K's performance per watt under a variety of workloads, it's a definite improvement.

Intel chips have traditionally performed well under idle conditions, which is where most PCs spend most of their time. AMD needs to catch up.

Temperatures

It goes without saying that temperature results are heavily dependent on the capabilities of your cooler, case airflow, and ambient temperature. For temperature testing, we aim to maintain a constant ambient temperature of 22 degrees Celsius.

27

This is a solid result for the 285K. The peak temperature doesn't look outstanding, but our cooler was nowhere near as stressed as it was when running a 13th or 14th Gen i9 at full tilt. With a good quality AIO, the 285K should not result in screaming fans. Well done, Intel!

Final Thoughts

The Intel Core Ultra 9 285K is a difficult chip to judge. In some ways, it's excellent. Its performance under heavily threaded workloads and performance per watt are highlights. The inclusion of an NPU - even if it is underutilized right now while we wait for the software ecosystem to mature - will come into its own in the months and years ahead.

But then there's that puzzling gaming performance...

Enthusiasts looking for the best gaming performance should sit this one out, at least for now. Arrow Lake feels underdone. In time, we expect to see some gaming performance gains, whether it's via Windows, improved thread scheduling, game engine awareness, or BIOS updates. But counter to that argument is that Intel has had a lot of experience with the fundamentally similar disaggregated Meteor Lake for a long time, and such tweaks have had ample time to be developed over the last year or so. Time will tell.

Intel's gaming headache will only worsen with the Ryzen 7 9800X3D announcement. Unless AMD stuffed up, it's hard not to imagine that chip becoming the gaming champ. Hopefully, Intel will get its gaming performance issues sorted ASAP.

27

The other drawback is the relative value on offer. Considering the cost of upgrading to the new platform on top of just the CPU, that money would go a long way to buying a high-end graphics card, which will deliver a better gaming improvement than a CPU and platform upgrade unless you're running a system that's many years old.

Gaming aside, there are many things to like. The efficiency gains are impressive, and the integrated graphics performance is a dramatic leap forward. True 1080p gaming is still a stretch, but if you add some upscaling into the mix, it's doable. It's a good option for those using a 285K for a bit of casual gaming on the side to play less demanding tiles such as popular Esport and MOBA games. Credit to Intel for the comprehensive capabilities of its Media Engine, too.

The new CPUs come with an impressive new platform. Though Z890 motherboards will be expensive to start with, they don't lack features. Intel users can finally run a PCIe 5.0 x4 SSD and x16 graphics card without compromise. Add to that features such as USB 4, Wi-Fi 7, and CUDIMM support, among other things. Though we have no confirmation from Intel, Z890 boards will likely support at least one more generation of CPUs.

We have many Z890 motherboards in hand, so stay tuned for reviews of those in the coming weeks.

So, should you spend your hard-earned money on a Core Ultra 9 285K? For anyone using chips from the 13th or 14th Generation or Ryzen 7000 families, we'd say no, with an asterisk. If your tasks can fully utilize lots of threads, and you don't want to spend money on a discrete graphics card, it's a solid option. And anyone coming from a system that's several years old will be impressed by the performance and efficiency gains on offer.

Intel's move to a disaggregated architecture represents its biggest architectural design shift in over a decade. Hopefully, it will serve as a stepping stone for future architectures like the original Zen 1 did for AMD. But for now, the 285K is a tough chip to recommend. 13th and 14th Gen chips retain their value, while AMD's Ryzen 7000 and 9000 series chips are highly competitive. Consider the 285K if you can fully utilize its strengths, but with unanswered questions over its spotty performance and ultimate potential, it's best to take a wait-and-see approach with the 285K and Arrow Lake.