AMD has quietly released its FSR 4.1 upscaling technology for Radeon RX 7000 graphics cards, fulfilling a promise made earlier this year. The catch? Early observations suggest the visual fidelity upgrade comes with an unexpected frame rate penalty on the older RDNA 3 architecture. As more gamers test the new driver, data points are accumulating: FSR 4.1 can be slower than its predecessor, FSR 3.1, even as it clears up shimmering, ghosting, and fine-detail loss.

The driver update—Adrenalin 25.3.2 as of this article’s research window—enables FSR 4.1 across the Radeon RX 7900 XTX, RX 7900 XT, RX 7800 XT, and other 7000-series cards. These GPUs were originally launched with support for FSR 3.1, an algorithmic upscaler that relied on hand-tuned filters and temporal feedback. FSR 4.1, by contrast, introduces a convolutional neural network (CNN) that runs on AMD’s AI accelerators. On RDNA 4 (the RX 9000 series) this hardware was designed to handle inference efficiently; on RDNA 3, the same accelerators are less numerous and clocked differently, leading to longer processing times per frame.

What’s New in FSR 4.1?

FSR 4.1 represents a fundamental shift in AMD’s upscaling philosophy. Instead of a purely analytical approach, it now uses machine learning to reconstruct a higher-resolution image from a lower-resolution input. This allows the algorithm to infer details that traditional algorithms miss, producing cleaner edges, more stable motion, and better handling of transparencies and thin geometry.

AMD first demonstrated FSR 4 during the RDNA 4 launch, showcasing side-by-side comparisons that highlighted reduced flicker on distant power lines and sharper foliage in titles like Horizon Forbidden West and Starfield. The 4.1 revision, which has now reached RX 7000 cards, adds refined model training and broader game integration—though the exact differences between FSR 4.0 and 4.1 remain subtle, focused on tail-case artifacts.

Performance Findings on RDNA 3

Independent testers across multiple forums and early YouTube benchmarks are reporting a consistent pattern: at the same internal resolution and quality preset, FSR 4.1 can cut frame rates by anywhere from 5% to 15% compared to FSR 3.1 on an RX 7900 XTX or 7800 XT. In extreme moments, such as heavy transparencies or rapid camera moves, the gap widens further, indicating that the AI cost scales with scene complexity.

For context, a gamer playing Cyberpunk 2077 at 4K with FSR 3.1 “Quality” might see 78 fps on a reference RX 7900 XTX. Switching to FSR 4.1 “Quality” under the same scene could drop the count to 67 fps. That’s a tangible decrease, though still within a playable range for many. The same test at 1440p often shows a slightly smaller relative hit because the GPU has fewer pixels to upscale, but the trend holds.

Why Is It Slower on RDNA 3?

The performance trade-off isn’t a bug—it stems from how FSR 4.1 offloads work to the GPU’s AI Matrix cores. On RDNA 3, AMD incorporated dedicated AI accelerators known as “AI Accelerators” within each compute unit, but they are less powerful than the 2nd-generation AI engines found in RDNA 4. Specifically, RDNA 3’s AI hardware delivers up to 2.7x lower AI throughput per cycle compared to RDNA 4, and the software stack may not yet be hand-optimized for the older silicon.

When a game engine hands a low-resolution frame to FSR 4.1, the CNN must process every pixel to hallucinate the missing detail. That inference pass takes a certain number of milliseconds. On RDNA 4, it’s fast enough to be cheaper than the hand-crafted FSR 3.1 pipeline. On RDNA 3, the computation takes longer, and because upscaling is still a serial step in the render chain, that extra time directly hits the overall frame time.

There’s also a difference in how the drivers schedule AI workloads on the two architectures. AMD’s engineering team will likely fine-tune this scheduling in future driver releases, so today’s performance numbers may not be final.

Image Quality Uplift

Despite the speed regression, the visual improvements are immediate and striking. Testers note a substantial reduction in the “fizzle” that FSR 3.1 exhibits on metallic surfaces and gratings. Fine text in in-game UI elements, such as HUD markers in racing games, is noticeably crisper. Thin objects—chainlink fences, power cables, character hair—appear with fewer temporal artifacts and less break-up during motion.

At lower internal resolutions (Performance or Ultra Performance modes), the quality gap widens. FSR 4.1’s AI reconstruction manages to hold together details that FSR 3.1 would smudge into an unrecognizable mush. This makes upscaling from 1080p to 4K more viable, especially on a 4K display where pixel peeping is common.

However, FSR 4.1 is not immune to its own artifacts. Some testers report a slight painterly effect in dense foliage or under heavy depth-of-field bokeh, likely a result of the CNN’s limited training data. Additionally, the AI model introduces a subtle latency overhead—though it’s typically under a millisecond and unlikely to be felt in all but the most competitive esports scenarios.

How to Enable FSR 4.1 on Your RX 7000 GPU

Enabling FSR 4.1 requires a compatible game, AMD’s Adrenalin Software version 25.3.2 or later, and an RDNA 3 graphics card. Not every title that supports FSR 3.1 automatically supports FSR 4.1; the game must be updated with the newer SDK or have its FSR integration patched. At launch, the list of supported games includes major AAA releases such as Marvel’s Spider-Man 2, God of War Ragnarök, The Last of Us Part I, and several others. AMD maintains an official list, which is growing weekly.

Within the Adrenalin control panel, FSR 4.1 can be overridden globally if the driver detects a game using FSR 3.1. Users should navigate to the “Gaming” tab, select “Graphics,” and expand the “FidelityFX Super Resolution” options. There, they can toggle between “Perform Automatic” (which lets the driver decide) and the specific FSR 4.1 mode. Some enthusiasts recommend per-game profiles because the performance impact is not uniform—a title gentle on the AI engine may show almost no penalty, while another could lose double-digit fps.

Community and Enthusiast Reaction

Since the driver’s quiet rollout, forums and social media have lit up with debate. The core disagreement echoes past transitions, like DLSS 3 frame generation: does the image quality uplift justify a performance hit? For gamers with high-refresh-rate monitors, the loss of 10 fps may mean slipping below a 120 Hz cap, making FSR 3.1 the pragmatic choice. For those playing slower-paced narrative games at 4K/60 Hz, the extra crispness is a pure win.

Some overclockers have already begun experimenting with raising the GPU’s base clock to offset the AI compute cost, but early results suggest that the bottleneck isn’t pure shader performance—it’s the fixed-function AI accelerators.

The Bigger Picture: Competition and Ecosystem

FSR 4.1’s arrival on RDNA 3 also reshapes AMD’s competitive stance against NVIDIA DLSS and Intel XeSS. DLSS has long held a quality advantage because it is AI-native and runs on dedicated Tensor cores. With FSR 4.1, AMD closes that gap significantly, often matching DLSS 3.8 in blind side-by-side comparisons, according to some early A/B tests. However, on NVIDIA hardware, DLSS has minimal performance overhead because the Tensor cores are separated from shader execution. On RDNA 3, the AI accelerators exist within the compute units and compete for register and cache resources, which may contribute to the slower speeds.

This nuance also matters for the future. As AMD continues to refine the FSR 4 model, it could become more performant on RDNA 3. The company’s physics-informed AI techniques and driver-level optimizations could shave milliseconds off the inference step. Developers may also implement hybrid modes that selectively apply the AI pass only to parts of the frame that benefit most.

Is the Upgrade Worth It?

The answer depends on your display, games, and tolerance for framerate dips. If you prize image clarity and often notice upscaling artifacts, FSR 4.1 on RDNA 3 is a compelling upgrade—the improvements are real, and AMD deserves credit for backporting the technology rather than locking it to new hardware. But if you’re a competitive gamer who counts every frame, or if you’re on a mid-range RX 7600 where every percentage point matters, you may want to stick with FSR 3.1 for now and wait for driver updates.

Ultimately, the move underscores a broader truth: AI upscaling, while powerful, is not a one-size-fits-all acceleration. As hardware generations diverge in AI capability, the software must be tuned carefully. For now, the Radeon RX 7000 series gains a powerful new tool—but one that its users will need to test themselves, game by game, to find the sweet spot between beauty and speed.

Looking Forward

AMD has not publicly committed to a specific timeline for performance improvements on RDNA 3, but a spokesperson noted that “we continue to work on optimizing the machine learning algorithms to better leverage the hardware in previous generations.” That suggests iterative driver updates are on the horizon. Meanwhile, the company is expected to bring FSR 4.1 to integrated RDNA 3 graphics in APUs, which could further broaden the user base and accelerate optimization.

For Windows gamers, the arrival of FSR 4.1 on older hardware is a positive sign—proof that AMD is willing to extend the lifespan of its GPUs with meaningful software features. And while the initial performance trade-off may sting, early adopters are already posting their before-and-after screenshots, and the consensus is clear: once you’ve seen the sharper image, it’s hard to go back.