AMD penned a blog post today to clarify Zen 5's gaming performance and address criticism in the wake of its turbulent Ryzen 9000 launch. Many reviews came to different conclusions about Zen 5’s gaming performance, and there were also large disparities between AMD's marketed performance values and those measured by reviewers. AMD contacted reviewers to investigate potential issues that led to the disparities and attributes much of the variance between reviews to disparate test methodologies, hardware, and software settings, which we’ll cover below. The company has also revised its internal benchmark suite with an updated list of game titles and shared revised performance projections.
As a result, AMD updated its gaming performance projections for Ryzen 9000, which it originally measured at an average of 6% faster than Intel, to now saying the processors are generally at parity in gaming performance when the Intel chips are tested with optimized settings. However, even after retesting, AMD has largely stuck with the same performance improvement projection for Ryzen 9000 chips over the prior-gen Ryzen 7000 models.
AMD also shared some of the performance improvements it has seen using an “Admin” Windows profile instead of the usual user account, which unlocks at least some additional performance through a branch prediction optimization. This feature will come to standard Windows 11 accounts via a Windows update.
We spoke with AMD about the updated testing and asked about an ongoing issue with its chipset drivers. The company assured us it would be corrected soon. We also asked the company about its stance on using in-built benchmarks versus custom gaming scenes.
AMD’s revised Zen 5 Ryzen 9000 performance projections against Intel
As is common with reviewers’ guides, AMD’s Ryzen 9000 documentation contained its own internal benchmark results for reviewers to use to determine whether their results were in line with expectations. AMD’s test results were different from those measured by reviewers, and AMD says that several factors contributed to those disparities.
The mix of game titles used in a test suite can greatly impact average performance metrics. AMD based its original suite of gaming benchmarks on a range of esports, AAA, and older games, but the suite trended towards older titles or games that weren’t as CPU-limited as newer fare. To generate a better view of modern game performance, AMD has added several new titles to its suite for its new benchmarks (list further below).
AMD says that Intel’s instability issues also contributed to the company’s original findings. Intel’s 13th- and 14th-Gen Core processors have suffered from an instability bug, leading Intel to advise its customers to use conservative power settings while they awaited a microcode patch to address the issues. As such, AMD’s lab generated its original comparison benchmarks using Intel’s baseline power profile. AMD wasn’t alone in making this adjustment — many reviewers (us included) used Intel’s more conservative power settings for the reviews (we’ve also since updated the reviews with results from Intel’s patch). AMD also used DDR5-6000 memory for its original tests for the Intel systems (the same memory settings it used for its Ryzen test system).
With these parameters, AMD says it originally found that its processors were, on average, 6% faster than Intel’s chips.
Intel released a microcode patch the day Ryzen 9000 reviews arrived, and, in tandem with testing with a broader suite of games, AMD generated its new projections using the more aggressive 'extreme' power delivery profile. Also, because Intel chips can support faster memory speeds, AMD switched to DDR5-7200 memory for Intel systems. As a result, AMD revised its performance data from a 6% lead (on average) over Intel’s systems to now saying the processors are at “parity in gaming using the most popular games included in the reviews.”
"When comparing to the competition using optimal settings, higher memory speed and extreme power delivery profile for the competition and Windows 11, version 24H2 for both (see details below), we see a double-digit lead for Ryzen 9000 Series in productivity and creator applications, ~30% lead in AI workloads, and parity in gaming using the most popular games included in the reviews," the blog reads.
It's noteworthy that while AMD originally cites that it claimed a 6% average, which encompasses all four of the Ryzen 9000 chips against their competitors, the company did cite an 11% gain for the flagship Ryzen 9 9950X over the Intel Core i9-14900K in its initial marketing materials. That implies that its gaming performance deltas varied by chip.
AMD’s Ryzen 9000 over Ryzen 7000 projections
AMD also made a minor change to its projections for performance gains with Ryzen 9000 over Ryzen 7000 — but not by much. The company originally claimed a generational improvement in 1080p gaming of 9%, but has now revised that projection to a 5-8% improvement. The company's use of a range of performance instead of an absolute number is noteworthy, implying this could vary based on the specific processors being compared.
Naturally, reviewers can see larger or smaller gains based on the totality of their test suite — the number and type of games chosen can greatly impact overall measurements. For instance, we measured a 12% overall gaming improvement for the 9600X over the previous-gen model, while some other reviewers saw single-percentage gains. (We’re also revisiting our testing with a broader set of game titles and a fresh test system image to double-check our findings.)
"On a generational basis, Ryzen 9000 Series delivers a ~10% improvement in productivity and creative workloads, ~25% improvement in AI workloads, and 5-8% improvement in gaming over the Ryzen 7000 Series," AMD states.
Admin account gives varying performance boosts
AMD generated its original and revised benchmarks using an Admin account instead of a standard user account. The Admin account ekes out a few extra percent of performance for Zen 5 processors by enabling certain branch prediction optimizations, which AMD says better leverage Zen 5’s wider branch prediction capacity. However, those optimizations aren’t active in standard user accounts.
AMD is working with Microsoft to deliver this ‘AMD-specific' branch prediction code for standard user accounts as an optional update (the feature is available now in the Preview Channel). AMD says this update typically improves performance by 2 to 3% overall. However, outliers can be quite large in some applications, as shown in the table below (game performance shown as average fps, 1080p high).
This updated code will also improve performance for Zen 4 and Zen 3 processors, but it has yet to be seen if Intel chips would also benefit (AMD does list it as ‘AMD-specific,’ but it’s unclear if there is also an Intel-specific version of the code).
The Impact of VBS
AMD cited several other contributing factors to the disparity between the differing results. Some reviewers use the Virtualized-Based Security (VBS) feature on their test bed, while others disable it. AMD tests with VBS active.
Microsoft enables VBS by default to strengthen security, but it hampers gaming performance (we measured a 5% drop). The impact on performance can vary by chip due to specific architectural enhancements that improve virtualization performance, like Mode Based Execution Control (MBEC), so testing with the setting either on or off impacts the results. It can also impact the relative positioning of the processors — if one processor handles virtualization code better than another, that advantage could be erased if VBS is disabled.
VBS can introduce other issues—some motherboards can enable VBS by default, while others can disable it by default (I’ve seen this with older BIOS versions on some Intel motherboards). If the reviewer is unaware, this could lead to mismatched VBS settings between Intel and AMD test systems, thus skewing benchmark results. (Microsoft has posted instructions for gamers who want to disable this setting if they prize gaming performance more than heightened security, so we test with VBS disabled.)
AMD will fix its chipset drivers
AMD’s blog doesn’t address a pressing issue with its chipset drivers, but I got clarification that the company will soon release a fix.
AMD's chipset driver uses a core parking technique to boost performance with some dual-CCD chips, like its new Ryzen 9 9950X and the top-tier Ryzen 7000X3D parts. However, this feature can’t be uninstalled from the operating system — once you’ve installed an AMD chip that uses the mechanism, you’re stuck with core parking for the life of your operating system. If you later install another processor with the same operating system, the feature will persist and can continue to park cores (potentially unbeknownst to the user), thus reducing performance with processors not designed to use the feature.
The only known good fix is to completely reinstall the operating system. This also creates problems for reviewers who test multiple processors on the same motherboard. While AMD didn’t touch on the topic in its blog, it’s conceivable that this issue contributed to the inconsistent results we’ve seen with the first wave of Ryzen 9000 reviews.
This issue has been known for some time, so I asked David McAfee, the Corporate VP and GM of AMD’s Client Channel Business, if there is a fix coming.
“[…] I will tell you it's in development. Our goal is to have that in your hands by the time we do our X3D reviews [Ryzen 9000X3D]. The timing didn't work out here, which was unfortunate, but I think this is a critical piece of ensuring consistent performance across a range of processor compares in the same socket,” McAfee responded.
In-built benchmarks versus custom benchmark scenes
Many of the latest titles have built-in benchmarks that reviewers use for testing, so measuring performance in games that don’t have an in-built benchmark requires finding a specific scene in the game and using it to measure performance. AMD’s original test suite used only in-built benchmarks, which has been fingered as a possible reason for the difference between AMD’s benchmarks and some reviews.
“Wherever possible, we are using built-in benchmarks, and if I'm completely honest with you, that's a lot of what has driven our game test suite for quite some time," McAfee said. "Having repeatable built-in game benchmarks is a far easier way to ensure that we're all speaking the same language, as opposed to just a title where everybody comes up with their own methodology on how to test, and then you're really talking apples and oranges and cherries.”
“Built-in benchmarks are by far our preferred approach. Anytime that we have an ISV that we work with and do dedicated development work on a title, we are always trying to get benchmarks integrated into those titles so that there's a standard way to produce results,” he concluded.
- Cyberpunk 2077
- Hitman 3
- Far Cry 6
- Shadow of the Tomb Raider
- Watch Dogs: Legion
- F1 2022
- Borderlands 3
- F1 2023
- Civilization VI
- Forza Horizon 5
- Starfield CPU
- Red Dead Redemption 2
- Warhammer 3
- DOOM Eternal
- CS 2
- Guardians of the Galaxy
- Rainbow Six Siege
- Horizon Zero Dawn
- Spider Man Remastered
Here's the list of games AMD used during its investigation. AMD has now added two game titles to its suite that don’t have in-built benchmarks, so it is obviously now incorporating at least some custom scenes. The remaining 17 titles have in-built benchmarks.
Thoughts
AMD's blog post seeks to clear the air with enthusiasts disappointed with the smaller-than-expected gaming performance increases with its new Zen 5 processors, but the effect remains to be seen. AMD has effectively dialed its comparison to Intel chips in gaming down to zero, saying that Zen 5 is at parity with Raptor Lake Refresh. Our benchmarks still show an advantage for Intel's higher-end chips over the flagship Ryzen 9 chips in gaming. Meanwhile, the Ryzen 5 and 7 chips are more competitive with Intel's price-comparable models, but Intel holds a lead. Naturally, that can vary by test suite.
AMD is also confident that Zen 5 processors deliver a gain over Intel's chips in productivity applications, and that Ryzen 9000 also delivers a 5-8% generational gain over the Ryzen 7000 series in gaming and other applications.
The Windows 11 patch appears at least moderately promising in some workloads, but as always, we'll have to wait for independent testing before passing judgment. In the meantime, we're working on our re-examination of Zen 5's gaming performance, which we hope to publish next week.
