Our testing shows that Intel’s fix for its Arrow Lake chips isn’t effective in addressing the chips’ lackluster gaming performance, at least on the motherboards we tested with. And we found that the Core Ultra 9 285K’s updated gaming performance with one motherboard is now slightly slower than before. Additionally, the required operating system update has improved gaming performance for the prior-gen Raptor Lake Refresh even more than the Arrow Lake chips, so the flagship Core Ultra 9 285K falls even further behind its predecessor. As you’ll see in our benchmarks below, the Core Ultra 9 285K still does not meet Intel’s initial gaming performance marketing claims and will not make our list of the best CPUs for gaming.
The Intel ‘Arrow Lake’ Core Ultra 200S launch was marred with lower gaming performance than the company promised, failing to match the prior-gen Core i9-14900K flagship in gaming despite the company’s already-uninspiring claims of parity. Intel pointed to numerous issues as the source of the low gaming performance and issued fixes via both Windows and BIOS patches.
At CES 2025, Intel issued its own benchmarks with claims of up to 25% more gaming performance with the ‘fix,’ a claim we’ve unfortunately seen parroted by the press. But those gains only apply to certain very specific issues that not all reviewers and users will have encountered. As we’ll cover in-depth below, there’s a copious amount of wiggle room in Intel’s dubious claims of performance gains, but our tests show the patches don’t have a tangible impact on the competitive positioning against AMD, and even worse, Arrow Lake now fares worse in gaming against its predecessor.
Testing Details
Intel’s ‘fix’ requires two basic components: Windows 11 build 26100.2314 (or newer) and microcode version 0x114 with CSME firmware kit 19.0.0.1854v2.2 (or newer). For our original review, we tested the then-current Windows version 26100.2033. We moved to version 26100.2605 for the configurations that represent patched performance. Intel says the newer version has an improved Power Performance Management (PPM) package that ensures good performance when using the balanced power profile. However, as with all reviewers who follow the most basic of best practices, we originally tested with the High Performance power profile for Intel systems, so we expect minimal to no impact on our testing from this change. You can read about the rest of the fixes here.
Cyberpunk 2077 had a rather large performance increase from a fix issued for the game code. However, Intel says this was an issue of the game dev’s own making, and the dev fixed the issue itself. Intel says we shouldn’t expect further game code updates that will boost Arrow Lake’s performance in the future.
Multiple game titles have also received updates since our original review, thus making it impossible to replicate those results with the newer test configs. To keep things apples-to-apples, we took our original test setup with the launch version of the BIOS and Windows for a spin back through our test suite to ensure that we don’t attribute performance gains from game code updates to a ‘fix’ from Intel. Then, we tested the impact after updating to the latest BIOS/Windows versions.
As such, the entries below marked with ‘Original’ represent the original launch BIOS and firmware, but have new updated testing to reflect the current state of the game code. The entries marked with ‘New FW-OS’ represent testing with the cumulative impacts of all updates. For the sake of expediency, we only tested with standard DDR5 memory (no CUDIMMS) and two motherboard platforms.
Core Ultra 9 285K gaming performance
Image 1 of 14
As you can see above, the Asus motherboard paired with the Core 9 285K actually sees a small performance regression in gaming after the patch – the unpatched 285K configuration is 3% slower than the newly-patched configuration. I retested this condition multiple times, and Asus has yet to respond to our queries on the matter.
We shifted gears to testing on the MSI motherboard to see if we could expect performance regressions with all motherboards. The MSI motherboard started from a much lower bar with the original firmware/OS, but it did make at least a decent 3.7% step forward. However, it still trails the original unpatched Asus configuration with the same setup we used for our review by 1.9%.
Perhaps more importantly, compared to the fastest patched 285K results on the MSI motherboard, the Ryzen 9 9950X is now 6.5% faster (it was ~3% faster in our original review), and the Ryzen 7 9800X3D remains nearly 40% faster than the 285K – it isn’t close. That means the fix has not altered Arrow Lake’s competitive positioning in a positive way versus AMD’s processors.
More concerning for Intel is that its previous-gen Core i9-14900K experienced much stronger uplift than the Core 9 285K from updating to the new version of Windows. We only updated the OS for the updated 14900K config – no new firmware had been released for our test motherboard since the 285K review. As you can see, the 14900K is now 7% faster than the testing with the older version of Windows. It appears that Windows has corrected some sort of issue with all Intel processors here, leading to the 14900K now being 14% faster than the 285K.
For reference, we originally measured the 14900K at 6.4% faster than the 285K in our launch day review, but now the 14900K is 14% faster than the updated 285K. Again, this trails Intel’s original performance claims of the 285K having parity with the 14900K.
So far in our game performance testing and the testing we’ve seen from other media outlets, while Intel has perhaps fixed a few corner cases, it surely has not fixed the mess created when it set expectations for the Core Ultra 9 285K unrealistically high. The 285K still does not live up to those expectations, and the fact of the matter is that the previous-gen Intel chips are demonstrably faster in gaming.
Core Ultra 9 285K productivity performance
Image 1 of 18
That’s not to say Arrow Lake is all bad, of course. The 285K has its charms in generational gains to single- and multi-threaded workloads, though AMD still holds the title in the latter category.
Here, we can see that the Core Ultra ‘fixes’ had no impact on overall performance in either our cumulative measurements of performance in single- or multi-threaded productivity workloads. We did see a slight regression for the 14900K with the updated OS, but only by a scant 0.8%, which falls within the expected variance.
Conclusion
Image 1 of 16
Here are Intel’s performance claims from a presentation during CES 2025 that outlines the impact of its patch. Several of these issues will not have impacted skilled reviewers, such as using a balanced power profile or not verifying that APO was working (APO impacts a limited number of games that most good reviewers test anyway). On the matter of APO, this was available to reviewers for launch-day reviews – we used it – and merely required one to simply verify it was working. One of the other examples shows an improvement in 7zip, but that has nothing to do with the shortcomings in gaming.
You’ll also notice that Intel includes improved performance in Cyberpunk 2077 in the summary slide. However, the company has also said the issue with that title was self-inflicted by the devs and fixed without Intel’s prodding. Yet it’s chalked up as an Intel win from the ‘fix.’ Intel also doesn’t mention that the 14900K also saw a solid boost from the updated Cyberpunk 2077 game code. The same can also be said about Far Cry 6 – the 14900K benefitted far more than the 285K from the move forward to the new version of Windows.
Intel’s presentation clearly shows the performance impact of the various features being turned on or off, but this is misleading, and Intel’s statements are vague and nebulous. You may or may not suffer from any or some of these issues, and the impact of each issue could vary greatly depending on your setup. All of the claims in the charts above, or none of the claims, or anything in between, could or could not apply to you and your system. Intel also says all of these issues could impact you in one moment in time but not in another, despite nothing (like settings) having been changed. Here’s how Intel puts it:
“The exact performance uplift you will experience with these updates depends on the specific issue, or combination of issues, present on your system when your data was originally collected. Results also depend on your selection of games or applications. Some issues are more elusive than others, more relevant to certain workload characteristics, and/or intermittent in nature.”
That makes pushing back definitively on the claims nearly impossible. Regardless, Intel’s claims of the performance being faster or slower with items toggled on or off still doesn’t clarify the correct issue – the 285K’s competitive positioning. You’ll notice that none of Intel’s tests above include competing processors, be they a previous-gen Intel part or any AMD part. That’s why you don’t see important things like the 14900K’s increased performance in Cyberpunk 2077 and Far Cry 6.
These tactics and the test results make this whole ‘fix’ exercise feel more like misdirection and spin than an actual fix. Yes, Intel admits that it failed to make sure that some issues were addressed in a way that would apply evenly across all users, and it has now corrected those issues. However, it still hasn’t brought the chips up to the level of performance it originally promised, and even if it merely matched its previous-gen chips in gaming as it claimed, that’s still not great. We expect generational improvements in performance, and anything less is rightly frowned upon.
At the end of the day, Intel’s fixes for its various failings did not demonstrably ‘fix’ the Core Ultra 9 285K’s gaming performance in any meaningful way, and they certainly aren’t enough to meet the company’s original marketing claims or change the competitive positioning of its lackluster Arrow Lake chips. In fact, it looks like Arrow Lake is moving backward. Despite its other positive attributes, the Core Ultra 285K simply isn’t the best option for gaming.
