AMD says FSR frame pacing is fixed. What the driver update actually promises
AMD shipped driver-assisted frame pacing for FSR frame generation in SDK 2.4 and Adrenalin 26.10.1, and it only reaches RX 9000 cards, with no independent slow motion verification yet.
Hero image: from Hardware Unboxed's slow-motion frame pacing investigation, published on YouTube.
In August, Hardware Unboxed showed that FSR 4 frame generation could post 120 FPS on screen while delivering motion that felt closer to 60. AMD spent two months without a public answer, then responded on October 20 with a promise of more consistent frame pacing. The mechanics behind that promise are now public, so it is worth separating what AMD committed to from what it left unsaid.
The measurement that started this
Hardware Unboxed filmed FSR 4 frame generation on a Radeon RX 9070 XT with a slow motion camera at 960 fps, running the driver available at the time, 26.7.1. The footage showed generated frames spending far less time on screen than rendered ones. In several games one frame stayed up roughly four times longer than the frame after it, so a 120 FPS output built on a 60 FPS base felt much rougher than the counter suggested.
The pattern held across the vast majority of games they tested, and no title came out clean. God of War Ragnarok, Cyberpunk 2077, Mafia: The Old Country, Assassin's Creed Black Flag Resynced, Resident Evil Requiem, Alan Wake 2 and Crimson Desert were the worst cases. Borderlands 4 and Lego Batman: Legacy of the Dark Knight came closest to even pacing, with problems still showing up in specific situations. Turning on V-Sync changed nothing. The channel also traced the same behavior back to FSR 3 frame generation, which shipped in 2023, so the flaw predates FSR 4 entirely.
What AMD promised, in its own words
At an AMD presentation on October 20, Jack Huynh, senior vice president and general manager of the Computing and Graphics Group, addressed the finding directly. His line, as quoted by outlets covering the event: "We'll also deliver new enhancements to FSR Frame Generation with more consistent frame pacing for smoother gameplay."
An accompanying company statement went further on delivery. AMD said it is "rolling out new frame pacing improvements for FSR Frame Generation on AMD Radeon RX 9000 Series graphics cards, delivering more consistent frame pacing for a smoother, more responsive gaming experience in supported titles."
Read those sentences closely and the limits are visible. The promise covers RX 9000 cards, in supported titles, with no numbers attached. Nothing in the driver release notes quantifies the gain, and AMD has published no per-game frametime measurements of its own.
What actually shipped
The mechanism landed in FSR SDK 2.4 and the Adrenalin 26.10.1 driver in late October. AMD calls it Driver-Assisted Frame Pacing, and it rides on FSR Frame Generation Swapchain 4.0.0. In the old design, the game's CPU code busy-waited to time each present call. The new design hands final presentation timing to the display driver, closer to the scanout pipeline. AMD claims smoother frame delivery and lower CPU overhead as a result.
The conditions, taken from AMD's SDK documentation, are narrow:
- RX 9000 series or newer
- Adrenalin 26.10.1 or later, with Windows 11 and Hardware-accelerated GPU Scheduling as the validated configuration
- A game built on Frame Generation Swapchain 4.0.0 or later
- Exclusive or borderless fullscreen covering the display, VSync off, presentation inside the display's VRR window
When the conditions are met, the driver enables pacing automatically. When they are not, the swapchain falls back to the older path, and on pre-RDNA 4 cards that fallback is the analytical frame generation swapchain 3.1.7.
The release notes stay quiet
Open the official Adrenalin 26.10.1 release notes and search for pacing. You will find FSR Upscaling 4.1 support extended to RX 6000 cards and Ryzen AI processors, new game support for Call of Duty: Modern Warfare 4 and others, and a list of fixed issues. Frame pacing appears nowhere in the document. The feature is documented on AMD's developer portal instead, because it depends on the SDK as much as the driver.
That split explains who gets the fix and when. A driver install alone changes nothing for a game you already own, unless the developer rebuilds it on the new swapchain.
Who benefits, and who waits
RX 9000 owners are the stated audience, and early checks lean positive. Follow-up testing reported in the days after release found native FSR 4 titles pacing more evenly on 26.10.1, while games still using FSR 3.1.5 or older SDK builds kept the jitter until developers update them. Community DLL swaps can drag some older titles onto the new path, though the same trick reportedly causes frame duplication in Cyberpunk 2077. These are secondhand summaries of video testing, and full per-game data remains thin.
RX 6000 and RX 7000 owners got part of the package. SDK 2.4 brings the ML-based FSR Upscaling 4.1 down to RX 6000, which is a real gain for image reconstruction on RDNA 2. The pacing fix is a different story. The SDK documentation restricts Driver-Assisted Frame Pacing to RX 9000 or newer, and AMD has said nothing about any equivalent for older architectures. Reporting from VideoCardz and Club386 flagged the same silence, and the question of whether RX 6000 and RX 7000 will ever see a pacing fix remains open.
What we have not measured
We have not reproduced the slow motion testing. We hold no frametime captures of the new driver, so there are no numbers from us in this article, and we will not invent any. Frame pacing is also hard to verify casually, since software counters misreport generated frames, which is exactly why Hardware Unboxed reached for a camera in the first place.
When we run our own check, the first question is whether the four-to-one frame duration spread actually shrinks in the worst offenders from the August list. The second is how many of the requirement hoops, fullscreen state and VSync settings among them, survive contact with real game menus. We will publish both with the camera method spelled out, scored against our methodology. Until that retest exists, treat AMD's claim as plausible and unverified, and judge frame generation by what your eyes see at your monitor.
