You know how all the hardware manufacturers are pushing laptop upgrades because you can't live without a 40 TOPS NPU for all the wunnerful AI things in Windows? Well, with its new generation of desktop CPUs, Intel backpedals on just how much NPU you really need -- and implicitly conveys how unimportant a Copilot Plus PC really is. The Core Ultra 200K overclockable processors are geared to gamers and creators (formerly codenamed Arrow Lake-S and now a subset of the Core Ultra 200S generation), similar to the preceding 14th-gen K-series Core CPUs. They'll be available from Oct. 24.
Intel also took the opportunity Thursday to confirm H and HX Core Ultra processors for gaming laptops would arrive before March 2025, likely meaning a CES 2025 announcement. Just like Intel does every year. Intel revealed the mobile and desktop Arrow Lake architecture underlying them during IFA in August.
| Core Ultra 9 285K | Core Ultra 7 265K | Core Ultra 5 245K | |
|---|---|---|---|
| P-Cores/ Max boost (GHz) | 8/4.6 | 8/5.5 | 6/5.2 |
| E cores | 16 | 12 | 8 |
| E core speed | 3.2/4.6 | 3.3/4.6 | 3.6/4.6 |
| E-cores/ Max boost (GHz) | 16/4.6 | 12/4.6 | 8/4.6 |
| total Threads | 24 | 20 | 14 |
| NPU cores/TOPS | 2 Gen 3/13 | 2 Gen 3/13 | 2 Gen 3/13 |
| Power class (base/boost) | 125W/250W | 125W/250W | 125W/159W |
| iGPU/max frequency (GHz) | 4 Xe-LPG cores/2.0 | 4 Xe-LPG cores/2.0 | 4 Xe-LPG cores/1.9 |
| Boxed price | $589 | $394 | $309 |
This announcement just seems odd. Arrow Lake is almost entirely an architecture overhaul generation: It doesn't promise any substantial speed improvements over its predecessor. Intel's goals were reconfiguring the chip to drop power consumption (and consequently heat) and to play catch-up on some features, all without suffering a performance hit. It does mean we can cram more powerful chips into more compact cases, which is a win. You may also notice that the number of cores now equals the number of threads: Time to say buh-bye to hyperthreading.
A few upgrades do make a dent in some aspects of performance and features, mostly because they're way overdue. There are changes to overclocking control, notably smaller clock increments, memory overclocking for new types (such as CUDIMM DDR5, which is necessary for 6400MT/s memory), the ability to bypass voltage limits if you're within certain thermal envelopes and more.
There's an upgraded integrated GPU based on last-gen Xe-LPG, which is faster than the sad version that lasted through 14th gen, and the upgrade was necessary for adding some of the engines that don't exist in the old UHD Graphics. They include the vector engines necessary for AI, compatibility with Intel's Arc software, support for Intel's XeSS upscaling technology and ray tracing plus other capabilities necessary for DX12 Ultimate. But it lacks the XMX instructions in the Lunar Lake iGPU which make XeSS faster, for example.
It adds an NPU, but it's the Core Ultra 100 series (Meteor Lake) laptop NPU 3, with fewer compute engines -- only two, for a total of 13 TOPS. And why has it taken so long to integrate Thunderbolt (TB4 in this case) into a desktop CPU? (That's at the expense of cheaper-to-implement USB4.)
This class of CPUs targets "enthusiast"-level gamers and creators, which means they'll mostly be in systems with separate graphics cards, and Intel's relying on those to make up any shortfalls in power, as it's done for a long time. (There are also KF variants for the Ultra 7 and Ultra 5, which ditch the iGPU and some commercial security features.) But that's no reason to start off with last-gen architecture.
Intel says the low-power NPU is enough to handle the processing for the camera, voice, interaction and more that Windows offloads to it and which makes systems with these chips "AI PCs." But either Intel doesn't think enthusiasts want the generic-but-more-complex AI capabilities that define a Copilot Plus PC or that Microsoft's Copilot Plus program doesn't matter for desktops, because it at least at the moment it doesn't allow for AI running on a discrete GPU. Or who knows -- maybe the epic fail that is Microsoft's Recall rollout makes everyone want to ignore a lot of the AI use cases.
Adding to the should-I-or-shouldn't-I conundrum, Arrow Lake CPUs aren't a pop-it-into-your-motherboard upgrade -- it uses a new socket with an updated 800 series supporting chipset, so you'll have to replace your motherboard. That makes it kind of a nonessential, wait a while choice. If you're buying or building a system, then it might makes sense, but I'd still hold off until we know how stable everything is. That's the downside of a new architecture rollout.