Intel Core Ultra 9 285, Core Ultra 7 265K, Core Ultra 5 245K and 235 in the post-test – Where Intel has really made gains, including a big surprise

1 - Introduction and Overview 2 - Gaming Performance 720p (1280 x 720 Pixels) 3 - Gaming Performance 1080p (1920 x 1080 Pixels) 4 - Gaming Performance 1440p (2560 x 1440 Pixels) 5 - Power Consumption and Efficiency 6 - Workstation Performance 7 - Summary and Conclusion

Since the market launch of the Core Ultra platform, I have consistently kept my entire test system in the then frozen configuration and also created a bit-identical SSD image to ensure later comparisons under identical starting conditions. For this retest, it was therefore sufficient to simply update the mainboard UEFI to the latest revision, apply all Windows updates and install the latest driver packages without changing a single line in the rest of the software environment. This makes it possible to see exactly what progress microcode, firmware and drivers have actually made since the launch.

This time, the focus is exclusively on three selected Intel CPUs, which were measured both in their original firmware version and in their current state. The AMD models tested in parallel at the time only serve as a static reference point in the diagrams, so that the time gap between then and now becomes more visible, as the actual topic remains the internal generational leap within the Core Ultra family. The fact that Ryzen processors also benefit from later optimizations thanks to the well-known fine-wine effect should be mentioned in passing, but is expressly not the subject of this analysis. In addition to the three Intel Core Ultra processors already included in the original launch test, I have now also included the Core Ultra 5 235 in the series of measurements for this follow-up test. This CPU is positioned below the Core Ultra 5 245K in terms of performance, but fits seamlessly into the platform from a technical perspective and rounds off the comparison at the bottom.

This time, I deliberately left out the Core Ultra 5 225F, which is even lower. The CPU is simply too slow for a serious comparison in a performance context, but it does have its own qualities, which I will discuss in a separate article. You already know at least the boxed cooler of this model from an earlier analysis, in which I also discussed its thermal behavior and the thermal paste. For today, however, the focus remains strictly on the models that can show a real leap in performance with the latest firmware and drivers.

It should be noted in advance that although the measured increase in average frame rates is there, it is not the real aha effect. The most surprising improvement is in the smoothness of the scrolling, which only becomes apparent in the detailed evaluation of the frame time distributions. However, we do not want to get ahead of ourselves at this point; all the relevant findings will follow in the further course of the article.

What has changed and what hasn’t?

Since the launch of the Core Ultra 200 series in October 2024, the interaction between motherboard firmware, Intel drivers and Windows 11 has changed almost monthly, with each sub-area contributing to the gradual elimination of the initial performance and stability problems. The firmware history of the MSI MEG Z980 Ace I used began with a boot UEFI that was still based on microcode 0x110 and CSME firmware 19.0.0.1762. The first corrections followed at the end of November in the form of a PR5 beta which, among other things, removed the risky DLVR bypass under microcode 0x112 and tightened the voltage limits for high load states, which increased stability but resulted in measurably higher memory latencies. The decisive step was the final UEFI 7E22v1A41 released by MSI at the turn of the year, which integrated the microcode 0x114 declared binding by Intel, including CSME 19.0.0.1854 v2.2, and should therefore enable a singular performance increase in the mid-single-digit percentage range in games and memory-intensive applications. On this basis, further revisions were successively delivered in spring 2025, which first enabled the factory activation of the new 200S boost profile and later also revised the memory training algorithm in version B3 so that CUDIMM modules up to DDR5-8000 could be booted reliably. A briefly distributed test build with microcode 0x118, on the other hand, did not bring any reproducible advantages and was reserved for beta users after internal measurements showed slight decreases in rare AVX2 workloads.

At the same time, Intel updated the driver stack several times. The launch version 32.0.101.6732 was the starting point, but the WHQL driver 32.0.101.6734 already increased the average frame rate on Lunar Lake handhelds by up to ten percent. The subsequent driver 32.0.101.6874, released in May, extended these optimizations to all Core Ultra 200V devices and at the same time increased the permissible iGPU memory usage from fifty to fifty-seven percent of the system RAM, allowing 9.12 GB to be reserved for the graphics unit in a 16 GB system, for example. In the latest iteration 32.0.101.6881 from June 9, several low-level errors in shader cache management were fixed and dynamic switching between balanced and performance profiles was permitted in the driver settings for the first time, which brings noticeable frame time smoothing, especially on systems with the Core Ultra 5-235. Intel has also consolidated the package of chipset, Wi-Fi and Bluetooth drivers so that installation is carried out via a standardized “one-click installer”, which also removes outdated power plans and switches to modern “energy-aware scheduling” guidelines.

On the operating system side, Microsoft delivered cumulative updates over the course of development cycles 23H2 and 24H2 that adapted the kernel scheduler to the heterogeneous core architecture of Arrow Lake and thus made the allocation of demanding threads to performance cores more reliable. With the current build 26100.4349 from June 10, 2025, further optimizations for DirectStorage pipelines and GPU hardware acceleration of AV1 encoding enabled by default were added, measurably improving load balancing between CPU, iGPU and SSD. This is another reason why I planned this article right now. WindowsCentral also points out in its current June report a reduction in background processes due to the now system-wide available Xbox Game Mode, which should enable lower frame latencies, especially on Intel iGPU systems like the Core Ultra 5 235. The optional 200S boost profile of the UEFI has been correctly recognized by Windows since build 26100.3775 and automatically integrated into the “High Performance” energy mode when XMP is activated, without the need for additional manual registry adjustments.

Under Linux, the same processors have been better addressed by a revised P-State governor since kernel 6.9. In night tests with microcode 0x114, Phoronix has recorded an average performance increase of around six percent compared to the initial situation, whereby the deviations between individual workloads are sometimes greater because the new governor is more precise when it comes to turbo limits. In addition to these scheduler adjustments, the Xe driver generation included in Mesa 25.1 contributes to the higher stability of the integrated Arc graphics, so that frame pacing outliers, which still frequently occurred at launch, have largely disappeared.

Overall, the system used here with the MEG Z980 Ace and the latest UEFI, the Intel driver 32.0.101.6881 and Windows 11 24H2 Build 26100.4349 results in a comprehensible, reproducible improvement compared to the launch state. For Linux users, the progress according to third-party tests is of a similar order of magnitude, with the new P-State-Governor now exploiting the potential of the performance cores almost as consistently as the current Windows scheduler. What has remained the same, however, is the existing hardware, the rest is new.

Test system and test methods remain the same

All relevant details about the test system used and the underlying software configuration have already been documented in detail in the originally published launch article, to which I refer you here. In order to avoid unnecessary repetition, I have deliberately refrained from listing the identical system information again in this post-test. The hardware components, BIOS settings and software versions mentioned there still form the basis, but have been supplemented by the latest firmware and driver versions, as described in the previous section.

Intel Core Ultra 9 285K, Core Ultra 7 265K and Core Ultra 5 245K Gaming and Workstation Review – When more suddenly becomes less