Intel's "Arrow Lake Refresh" Core Ultra 300 Series Comes with K and KF SKUs Only

[AleksandarK]

Back in February, we reported on Intel considering a resurrection of the "Arrow Lake Refresh" Core Ultra 300 series of CPUs. However, it seems that @Jaykihn, a reliable source of Intel leaks, has confirmed that Intel will indeed push the refresh of Arrow Lake in the form of Core Ultra 300 series of CPUs, limited to K and KF SKUs only. This means we will likely see only overclockable SKUs being refreshed, with possibly enhanced boost frequency and/or core count. With 125 W TDP, these new SKUs would target high-end markets, passionate overclockers, and system integrators selling high-end builds with these CPUs. The refresh's journey to confirmation has been turbulent. Initial rumors in 2023 suggested an ambitious core-count bump for Arrow Lake-S, followed by speculation that Intel would prioritize NPU upgrades, potentially mirroring Lunar Lake's reported 48 TOPS capability—a massive leap from the current Core Ultra 9 285 K's 13 TOPS. By late 2023, leaks hinted at the project's cancellation, but insiders like Chinese leaker Golden Pig Upgrade later revived hopes, asserting that desktop-focused "ARL-S Refresh" had been quietly resurrected.

Jaykihn's latest intel narrows the scope: only K/KF-series chips will receive tweaks, leaving non-overclockable SKUs untouched. Intel is also reportedly developing new performance profiles for existing Arrow Lake CPUs, separate from warranty-voiding BIOS tweaks. These optimizations, alongside the upcoming Intel Performance Optimization (IPO) program, aim to help OEMs and system integrators push pre-built systems further via safer, stability-focused adjustments to clock speeds, power limits, and memory overclocking. While enthusiasts may still prefer manual overclocking, IPO could democratize performance gains for mainstream users. The Core Ultra 300 series will slot into Intel's LGA-1851 roadmap between the base Arrow Lake-S (Core Ultra 200) and 2026's next-gen Nova Lake-S (Core Ultra 400).




This positions the refresh as a strategic stopgap, particularly with Intel's mobile division advancing toward Panther Lake (late 2025) and Nova Lake (2026). The desktop segment, however, risks being underrepresented until Nova Lake arrives, making Arrow Lake Refresh critical for sustaining momentum in the high-end DIY and AI PC markets. While the exact launch window remains unconfirmed, a late 2025 debut is expected. If the NPU upgrades materialize, the refresh could also serve as a testing ground for Intel's AI ambitions, bridging the gap between Lunar Lake's mobile-focused AI prowess and future desktop innovations. For now, the Core Ultra 300 series appears tailored to overclockers and filling roadmap gaps—proof that even a "limited" refresh has a lot of significance.

View at TechPowerUp Main Site | Source

 

[Zach_01]

Oh good… “new” 250+W SKUs for the enthusiasts.
We all know how good this was turn out to be with 14th gen.

 

[Ed_1]

If they can do similar optimizations they did from Alderlake to Raperlake it will be worthwhile. raise ring, D2D and NGU and whatever core freq while keeping voltage in check.

 

[Denver]

I assumed Intel had learned by now that pushing CPUs to the brink and labeling them as "new" products isn’t the smartest strategy. But hey, if they insist, go ahead: keep burning your brand to the ground.

 

[usiname]

If they can do similar optimizations they did from Alderlake to Raperlake it will be worthwhile. raise ring, D2D and NGU and whatever core freq while keeping voltage in check.

More likely it will be like Raptor > Raptor Refresh

 

[TheinsanegamerN]

Oh good… “new” 250+W SKUs for the enthusiasts.
We all know how good this was turn out to be with 14th gen.

Power usage was not the 14th gen's core problem. The issue was a design fault from the change to increase cache in the 13th gen. 12th gen parts can run at 400+ watt OCs for years without issue.

 

[Prima.Vera]

Unless they go with the new 18A or even lower with very good wields, this will be another DOA CPU like the previous series.

 

[SL2]

I have zero interest in buying this, but I'm actually curious if they can make a substantial difference compared to the current ones.
(No, not having expectations. You don't have to lecture me lol.)

Then again, if it was that good they'd refresh the whole lineup, or at least more of it. Or, is it a decision based on the infinite amount of CUltra 2 chips waiting to be sold?

 

[bug]

More likely it will be like Raptor > Raptor Refresh

Raptor Lake was the last in a long line of refreshes/incremental improvements. Arrow Lake is new architecture, so there's more potential gains to be had.
I say there's a good chance this will be better than Raptor Lake -> Raptor Lake refresh. But probably not by much, it's still a refresh. (Fwiw, Zen -> Zen+ was also a refresh, we're not looking at those kind of gains here.)

 

[phanbuey]

I really hope they're able to lower latency -- if so then the gains will come. The core itself is very strong, there were some really questionable design choices (like not having adamantium cache) and putting the memory controller on the far side of the chip, that I still can't quite understand.

 

[dj-electric]

Absolute despair. I like Z890 over X870E, but i wish we could do a magic swap of connectivity.

 

[ir_cow]

So 13900K/14900K on the 1851 socket because they couldn't put the Ultra 200 on the 1700??? Yes lets make things even more complicated to consumers by rebranding old tech into new.

 

[Onasi]

So 13900K/14900K on the 1851 socket because they couldn't put the Ultra 200 on the 1700??? Yes lets make things even more complicated to consumers by rebranding old tech into new.

Not sure you got this correct - it’s just the Ultra 200 re-branded as Ultra 300 and pumped up on frequency and, presumably, power. That’s it.

 

[kondamin]

Leftover n3b capacity as original skus aren’t selling and they need to keep the system,running not to get tsmc fines is my guess.

 

[db87]

Can we just ditch all the potato cores and make something proper for once?

 

[bug]

Can we just ditch all the potato cores and make something proper for once?

They're not "potato" by any metric. Love or hate E-cores, they let Intel run circles around AMD in heavily multithreaded scenarios.