Conclusion

Intel's new Core Ultra Arrow Lake-S desktop processors are a fantastic piece of engineering, and they are Intel's first chiplet-based desktop processors—AMD has been using chiplets for many years now, and it's the foundation for their tremendous success. While Intel managed to improve energy efficiency massively, Arrow Lake fell short of performance expectations—not only in our review, but across all reviewers. But this was back in October 2024 in our launch review. What's changed in the last 50-odd days? The company took stock of the press coverage of the processor, particularly on how it failed in delivering the performance gains highlighted in Intel's pre-launch press-decks. In the pre-launch brief, Intel spoke of gaming performance of the 285K being on-par with the Core i9-14900K, while the application performance and energy efficiency get ahead. The latter two claims mostly hold, but the gaming performance saw a performance regression—something unheard of for Intel on the desktop platform.

The investigation
As part of its investigation, Intel identified a number of reasons, and posted a situation report yesterday, on December 18. This prompted us to work on this mini-article to get a feel whether Arrow Lake is truly fixed. According to Intel, the first set of issues was OS-level scheduler problems caused by faulty collaborative processor power management (PPM), which cost the processor anywhere between 6% and 30% in performance. This was solved in a Windows 11 24H2 November 2024 update, which we installed for testing. But we didn't just test Windows 11 24H2 pre-update vs post-update, but also compared the performance to Windows 11 23H2. Our launch reviews used 23H2 for this exact reason—numbers on 24H2 looked fishy, so I opted for the older, but battle-tested 23H2.

Our test setup
In today's article we benched five configurations: 23H2 from October, 23H2 with the newest updates till today, and also added the newest ASUS BIOS into the mix, which includes the 0x113 microcode (not 0x114 yet). Then, we moved on to test Windows 11 24H2—first with a fresh installation that had no chance to update, so it lacks the KB5044384 package that includes Intel's fixes to the processor's power management profile. Finally, we updated Windows 11 24H2 with the newest everything, including KB5044384 and all other patches released this month.

Just to clarify, in addition to "23H2 with new BIOS", all Windows 11 24H2 green bar tests were done with the 1101 BIOS, too, which includes microcode 0x113. We updated this article on Dec 20th with the purple bar results, which are Windows 11 24H2 with the brand-new 1203 BIOS, which includes microcode 0x114.

Application results
Our application results confirm that Intel has definitely improved things with their PPM update. Especially when comparing 24H2 only, the differences are noteworthy and bring much-needed performance improvements. However, compared to Windows 11 23H2, we see no meaningful advancements overall. It is good to see 20% gains in our Avast antivirus test, which confirms that Intel's "up to 30%" isn't a completely random number. However, the whole claim is "6-30% (estimated)", which would suggest you'll see at least 6% in a majority of workloads—this is definitely NOT the case. Overall, the 285K cannot gain enough performance to make it a clear winner vs AMD's Ryzen 9000 lineup, or even 14900K Raptor Lake (it improves power consumption considerably though).

Gaming results
For our gaming benchmarks, we pulled out our humble GeForce RTX 4090 graphics card, which brings out the best in processors by making sure games are bottlenecked at the CPU, rather than the GPU. All game tests were done at 1080p with max settings. Here, too, we see substantial gains when comparing 24H2 only, but 23H2 still remains slightly faster. While it's close enough most of the time I do wonder how the performance profile of 24H2 can be so different on Arrow Lake. Looking at the big picture, the 285K flagship remains slower than even the previous generation mid-range chip, the Core i5-14600K.

"Elden Ring" is an interesting test, where both 24H2 December 2024 and Fresh Install configurations test at least 10% slower than the other configurations that run some form of 23H2. "Spider-Man Remastered" is where Arrow Lake shines, and gets ahead of the Core i9-14900K in all software configurations. It's faster than even the 7800X3D in this game. The only chip to beat it is the 9800X3D. All configurations are clumped up between the 14900K and 9800X3D, and 23H2 with October 2024 patch is the fastest configuration. The 285K remains the worst chip to have in our comparison for "Starfield" and "The Last of Us"—I don't think this is what anyone expected after Intel's announcement yesterday.

It seems that Intel did not "fix Arrow Lake," but they "fixed Arrow Lake on 24H2". No doubt, fixing your product's performance in the newest OS release specifically is important and a solid improvement—why isn't the company transparent and makes that clear in their messaging?

The new 0x114 Microcode
All eyes are now on the 0x114 Microcode Update Intel is planning to roll out in January 2025. The update will be encapsulated into motherboard UEFI firmware updates, and your motherboard vendor or desktop PC OEM should have these out around that time. Unlike the other OS-level patches Intel's managed to put out till now, the microcode is expected to broadly improve performance of Arrow Lake in games. Intel has done an internal run of around 35 game tests, where the microcode averaged a "single digit percentage gaming performance enhancement", according to the written statement. In a YouTube interview with HotHardware, Robert Hallock mentioned "3-8%."—let's hope these gains materialize and can help to boost Arrow Lake to competitive levels.

Update Dec 20th:
Last night, ASUS has released a new BIOS version (1203), which includes the new 0x114 microcode update that was promised for January. We tested the new version and are happy to report that it brings some additional, small, performance improvements in games and some applications. Unfortunately, Avast antivirus does lose some performance with this update. Overall, these small gains are appreciated. However, they do not change the overall positioning of Arrow Lake vs AMD Ryzen and Raptor Lake. Still, they are good news for people sticking with 23H2. Looking at the results, I'd say that with the new microcode, 24H2 is at least as good as 23H2 (for Arrow Lake performance specifically).

We also tested a new Intel ME Engine firmware update, by request of our forum members. This release isn't official yet, and I'm not sure if I can recommend it for general usage. Differences are tiny, some up, some down, and there's a (small) performance regression in Cinebench nT.

As of today, the Core Ultra 9 285K remains hard to recommend to PC gamers when looking at Intel's own 14th Gen Core Raptor Lake, and AMD's efficient Zen 5 X3D chips.