AMD FSR Redstone is launching today, and it’s hard not to feel a bit let down. AMD is calling this a "new era of gaming innovation," but the flagship Frame Generation feature is only partially functional, and every major addition is limited to RDNA 4 GPUs. In this article we’ll mostly focus on the state of Radeon Frame Generation, but if you’ve been hoping for a sweeping upgrade that brings AMD’s technology in line with Nvidia’s, we don’t have particularly good news for you.
What is AMD FSR Redstone?
AMD Redstone is a collection of AMD technologies under the FSR umbrella. You’ve already seen FSR 4 upscaling, which arrived earlier this year – now rebranded as simply "FSR Upscaling," with AMD dropping the version number e…
AMD FSR Redstone is launching today, and it’s hard not to feel a bit let down. AMD is calling this a "new era of gaming innovation," but the flagship Frame Generation feature is only partially functional, and every major addition is limited to RDNA 4 GPUs. In this article we’ll mostly focus on the state of Radeon Frame Generation, but if you’ve been hoping for a sweeping upgrade that brings AMD’s technology in line with Nvidia’s, we don’t have particularly good news for you.
What is AMD FSR Redstone?
AMD Redstone is a collection of AMD technologies under the FSR umbrella. You’ve already seen FSR 4 upscaling, which arrived earlier this year – now rebranded as simply "FSR Upscaling," with AMD dropping the version number entirely. There’s no update to the technology itself; this is still FSR 4, just wearing Redstone’s new label.
The real changes start with FSR Frame Generation, now powered by a machine-learning model that promises sharper, more stable frames. There’s also FSR Ray Regeneration, a neural-network-based ray tracing denoiser meant to play in the same space as Nvidia’s Ray Reconstruction. And then there’s FSR Radiance Caching, which uses a neural network to store and reinterpret lighting data to accelerate ray-traced scenes.
But for anyone on older Radeon hardware, the story gets bleak quickly. All of Redstone’s core features are exclusive to the RX 9000 series and the RDNA 4 architecture, including the AI-powered versions of Upscaling and Frame Generation. Rumors hinted that Redstone might eventually expand to older architectures, but that isn’t happening. Official support starts with RDNA 4 and moves forward from there.
Some features don’t offer fallback paths at all. FSR Radiance Caching and Ray Regeneration aren’t available in any form on RDNA 3 or earlier. Upscaling and Frame Generation do have fallback options, but temper your expectations – this is just the existing FSR 3 "Analytical" model, not a new or improved version. In other words, RDNA 1 through RDNA 3.5 users stay on the current FSR 3 track, while RDNA 4 gets the new machine-learning suite.
The most frustrating omission is FSR 4 INT8. We know it exists: AMD released the source code, we tested a compiled version, and it ran impressively well on RDNA 3 and 2 – exactly the kind of fallback you’d want given the lack of FP8 support on those architectures. But AMD isn’t including it as part of Redstone. The fallback for FSR Upscaling remains the older FSR 3 analytical model, and AMD hasn’t said whether INT8 upscaling will ever be officially released.
Still, for those on the latest Radeon hardware, it brings multiple new ML-driven features. Ray Regeneration has technically been available for a bit: it’s already live in Call of Duty: Black Ops 7 multiplayer – and the Redstone launch doesn’t add anything new to it. Black Ops 7 is the official debut title, and the feature isn’t in other games yet, so if you’ve seen footage of it already, you know what’s coming. We’ll dig into it ourselves soon.
FSR Radiance Caching, meanwhile, isn’t ready for testing at all. AMD says it won’t arrive until 2026. That leaves the new ML-powered FSR Frame Generation as Redstone’s biggest launch-day feature. Much like FSR 4 Upscaling, the latest driver includes a toggle that automatically upgrades FSR 3.1.4 Frame Generation to the new ML-based implementation.
As long as a game has integrated FSR 3.1.4 or later, the toggle appears, giving us roughly 30 supported titles at launch – big names like Cyberpunk 2077, Black Myth: Wukong, Hogwarts Legacy, and Black Ops 7. Developers will eventually be able to integrate the Redstone version natively, but at launch every supported title relies on the driver upgrade system.
Frame Pacing Issues
The problem with FSR Redstone Frame Generation is that it doesn’t work very well. AMD’s frame generation tech still suffers from poor pacing, something we flagged two years ago when FSR 3 first launched. As far as we can tell, this remains a problem in FSR Redstone, and we don’t believe this update has improved how frames are delivered to the display.
The easiest way to examine frame pacing is to look at a real-world output on a monitor filmed with a slow-motion camera. This shows how Frame Generation delivers frames to the display with adaptive sync (FreeSync) enabled, which is the default and recommended setup for most players.
Direct captures through Radeon Software don’t reveal how frames are actually presented, and capture cards with fixed refresh rates often mask the issue as well. To properly evaluate real-world pacing, we recorded footage at 960 FPS.
For a better representation of image quality, check out the HUB video below:
First, let’s look at Black Ops 7. With Frame Generation off, we get the default experience. We configured the game to run at about 60 FPS at 4K, and what you’ll see is a consistent delivery of frames. The frame rate does fluctuate slightly around that 60 FPS mark, but with adaptive sync enabled the interval between frames is consistent and this leads to a smooth output.
When we enable FSR Redstone Frame Generation, the output becomes choppy and juddery. Instead of each frame being perfectly paced, one frame is quickly followed by another frame on the display, creating a ‘double tap’ effect where consecutive frames have alternating short and long intervals. You get two frames quickly, then a longer pause, another two quick frames, a longer pause. Occasionally FSR Frame Generation seems to be able to create a smoother and more consistent output, but throughout most of this test run we see poor pacing. And it’s not limited to quick camera movement, in the video above we show an example of the same judder simply walking forward with Frame Generation enabled.
For a better representation of image quality, check out the HUB video below:
Now let’s compare DLSS Frame Generation in the same game, swapping the Radeon RX 9070 XT for a GeForce RTX 5070 Ti. In the same in-game location, Nvidia’s output is noticeably smoother and more evenly paced, resulting in a much better experience. The judder seen on the Radeon system is visible in real gameplay even though this footage was captured at 960 FPS; the GeForce output is clearly superior.
The issues aren’t limited to this one game. In God of War Ragnarok using FSR Frame Generation, again showing a juddery double-tap pattern. Whether V-Sync is on or off makes no difference to pacing; in these examples, V-Sync is off for both FSR and DLSS.
In Hogwarts Legacy we observed the same pacing problem, followed by the Nvidia version using DLSS Frame Generation which looks fine. In GTA V Enhanced Edition, you’ll even notice some screen tearing in the FSR output despite FreeSync being active. At identical settings, the DLSS version doesn’t exhibit these issues.
F1 25 shows similar behavior – subtle at higher frame rates, but visible when the footage is slowed down. And again, DLSS Frame Generation provides smoother, properly spaced output. Cyberpunk 2077 shows the same trend: inconsistent pacing with FSR Frame Generation and smoother delivery with DLSS.
For a better representation of image quality, check out the HUB video below:
Not every game shows the issue though. Mafia: The Old Country is more stable with FSR Frame Generation, producing output similar to DLSS.
Black Myth: Wukong isn’t too bad either, some minor choppiness, but much better than the worst offenders. However, after testing 10 supported titles with FSR Redstone Frame Generation, we found pacing issues in the majority of them, while those same games using DLSS Frame Generation were smoother and more evenly paced.
As a sanity check, we ran FSR Frame Generation on a completely different system and setup. After all, it could have been something specific to our main test machine. So we installed a different GPU (the Radeon RX 9060 XT) into a separate system with no shared components and connected it to a different display. The same problem appeared in Cyberpunk 2077 and F1 25, confirming that this is not isolated to our main test system.
Not Ready for Prime Time
The issue seems fairly straightforward: FSR Frame Generation is not properly delaying the output of generated frames. Consider an example where the game renders at 100 FPS, and Frame Generation pushes the final output to 200 FPS. The GPU renders a frame every 10 ms and generates an additional frame between those, ideally producing a new frame every 5 ms. For proper pacing, the system should wait 5 ms after displaying the rendered frame before showing the generated frame, then wait another 5 ms before the next rendered frame. That’s how you achieve smooth pacing, and it appears to be how Nvidia handles the process. At 200 FPS, each frame lands roughly 5 ms apart.
But if the delay is incorrect – or if the generated frame isn’t delayed at all – the output becomes choppy. Generated frames take far less time to create than rendered frames, so displaying both immediately causes a severe timing mismatch.
Once we discovered this issue early in testing, our enthusiasm for further testing of FSR Redstone Frame Generation quickly evaporated.
Using the same 100 FPS example, if the generated frame arrives 1 ms after the rendered frame, the system then waits around 9 ms for the next rendered frame. This creates irregular spacing between frames. Combine that with an adaptive sync monitor which shows the next frame as soon as it’s ready, and you get visible judder and potential tearing. V-Sync doesn’t solve this, as adaptive sync still prioritizes showing frames immediately.
This inconsistent pacing is especially noticeable on high-refresh-rate displays... the exact monitors that should benefit most from Frame Generation. Faster refresh rates reveal uneven frame delivery more clearly, amplifying judder. Lower-refresh displays naturally have longer gaps between frames, which can mask some pacing issues but increase the chance of tearing. While we did most of our testing on a 4K 240 Hz monitor, the pacing problems became even more obvious on a 1440p 540 Hz display.
We don’t know how FSR Frame Generation currently handles pacing internally. With FSR 3, AMD used some unusual methods, and the current behavior suggests a similar approach, one that results in insufficient frame delays and uneven pacing.
AMD is aware of our findings. We received access to FSR Redstone last week, emailed them about the pacing problems on Saturday, and they acknowledged the report, though they have not yet provided their own conclusions or potential fixes. Because these issues have persisted since FSR 3, we would be surprised if there were a simple solution. At this point, we believe the problem is inherent to how FSR Frame Generation functions.
The whole goal of Frame Generation is to improve the visual smoothness of the game output. It doesn’t boost performance or reduce latency, but it can create a more fluid output with less motion blur and greater clarity. That benefit disappears entirely if the generated output is choppy, the exact opposite of smooth.
Once we discovered this issue early in testing, our enthusiasm for further testing of FSR Redstone Frame Generation quickly evaporated.
Frame Generation Image Quality
We also wanted to briefly touch on the image quality of the technology, in case AMD is able to resolve the frame pacing problems and make the feature properly usable. Here’s Mafia: The Old Country running at 4K 120 FPS – so 60-to-120 FPS frame generation – using Quality upscaling from both brands. You might be wondering how we captured this footage without encountering pacing issues. That’s because using a fixed refresh rate instead of adaptive sync, locking the frame rate to the monitor’s refresh rate, and forcing V-Sync on at the driver level can stabilize the output. However, this is not a practical way to play games and may not even be achievable on some setups, so this is strictly a test configuration...
For a better representation of image quality comparisons, check out the HUB video below:
As with previous frame generation investigations, we’re playing back this footage at half speed so you can clearly see each frame. What stands out immediately in the FSR 3.1 footage is flicker in the car’s shadow.
Redstone Frame Generation significantly reduces this flicker, producing a far more stable shadow that aligns with the quality of DLSS Frame Gen. In real-time gameplay at 120 FPS, the FSR 3.1 flicker is extremely noticeable and distracting, so Redstone’s improvement is welcome. There is still some distortion every second frame in detailed shadow regions such as tire spokes, but this is much harder to detect than broader shadow instability.
We also noticed in Mafia that FSR Redstone handles trees in motion very well, while those elements appear more garbled with DLSS Frame Gen, giving AMD a clear win in this instance. Grass is blurrier with FSR, but frame-to-frame consistency is better than DLSS, suggesting that FSR’s interpolation is behaving more accurately.
For a better representation of image quality comparisons, check out the HUB video below:
In God of War Ragnarok, FSR and DLSS trade blows visually depending on the scene. In a first example, textures on Kratos’ back are more stable with FSR, while DLSS can look slightly distorted in motion. But in the next scene, DLSS shows better stability around the axe as it moves, and in some frames Redstone is actually more garbled than FSR 3.1, which is interesting.
In F1 25, the comparison is also mixed. DLSS Frame Gen does a better job preserving ground textures, while every second frame in the FSR Redstone output shows noticeable distortion in the road detail. As seen previously in Mafia, FSR 3.1 struggles with shadow stability, which Redstone fixes.
The newer ML-based FSR Frame Generation also handles fine details such as the antenna and pitot tubes at the front of the vehicle more effectively, with less ghosting than DLSS. DLSS additionally has trouble rendering the UI, since Nvidia’s algorithm applies frame generation to UI elements, while FSR ignores the UI and renders it at half rate, which is better at preserving details.
For a better representation of image quality comparisons, check out the HUB video below:
In Call of Duty: Black Ops 7, the technologies trade blows again. FSR 3.1 has the most flicker and instability in effects such as blood and particles, and Redstone improves this significantly. In other areas, the cleaner output varies on a frame-by-frame basis, but overall AMD’s technology gets reasonably close to DLSS.
Lastly, a quick look at Cyberpunk 2077. This is the weakest showing for FSR Redstone. The visual output appears broken in this game, with heavy blur and grain that we did not see in the other five titles. DLSS Frame Gen is clearly superior here, offering better stability and reduced blur.
FSR 3.1 isn’t great either, so our best guess is that this is simply a poor implementation of AMD’s frame generation technology in this game. It’s also a different engine from the rest, so it’s possible that FSR Frame Generation doesn’t work particularly well with this engine.
Frame Generation FPS Performance
As for output frame rate, FSR Redstone Frame Generation is not substantially heavier to run than the regular FSR 3.1 Frame Generation, now referred to as the "Analytical" version.
In Mafia: The Old Country, enabling Redstone reduced the output FPS by only a few frames. In F1 25, with output above 250 FPS, Redstone lowered performance by about 3%, which isn’t especially significant, and in God of War Ragnarok Redstone actually ran slightly faster than the Analytical model.
In all of these cases, FSR Frame Generation carries slightly less overhead than DLSS Frame Generation on similar hardware. In Mafia using Redstone, we saw a 69% increase in output frame rate on the RX 9070 XT compared to frame generation being off, while on the RTX 5070 Ti we saw a 61% boost using DLSS Frame Generation.
In F1 25, Radeon hardware saw a 40% increase versus 38% on GeForce, and in God of War Ragnarok we recorded a 72% boost with Redstone compared to 59% using DLSS.
A Rough Launch
Ultimately, FSR Redstone Frame Generation with the new ML model is not very impressive. This is disappointing because there are underlying strengths like image quality improvements compared to the older model, and AMD achieved these gains with minimal additional overhead. But frame pacing is broken for actual gameplay, which completely undermines the purpose of enabling frame generation at all.
On a typical adaptive sync monitor, the FSR Frame Generation output is choppy and inconsistent in most games, something that becomes especially noticeable during camera pans or fast movement.
AMD’s own promo video for FSR Redstone paints a much more positive picture
Making matters worse, it doesn’t seem like AMD has made meaningful progress in frame pacing since the original launch of FSR 3 back in 2023. We spent a lot of time in our FSR 3 analysis detailing how pacing was broken, and the issue is even clearer now that we can record better slow-motion footage. Smooth pacing is technically possible, it happens in certain games under certain conditions, but it doesn’t happen reliably across the broader sample of titles we tested with Redstone.
This is a significant problem for AMD because Nvidia’s DLSS Frame Generation is consistently smooth in the same games. We weren’t able to find an example where DLSS Frame Generation produced noticeable judder. DLSS "just works," while we tested numerous FSR configurations across multiple systems and none delivered consistent pacing without resorting to completely locked-down frame rate and refresh rate settings. Maybe there’s a hidden setup that resolves FSR’s pacing issues, but we haven’t found it, and AMD has yet to provide a solution.
On a more positive note, FSR Redstone’s image quality is very competitive with DLSS Frame Generation. The output isn’t identical and both approaches have visible artifacts (frame gen inherently produces them), but Redstone is a substantial improvement over FSR 3.1. There’s reduced shadow distortion, better handling of disocclusion, and a more stable image overall.
However, the quality of frame generation varies significantly between FSR Redstone and DLSS depending on the game or even the scene. In one moment you may see artifacts in the DLSS output that aren’t present in FSR, but switch to another scene and the situation reverses. Because generated frames only appear half the time, the differences between DLSS and FSR aren’t a major issue as long as there are no serious visual flaws, and Redstone generally avoids those.
Based on everything we’ve seen, if you’re interested in Frame Generation today, our recommendation is to buy a GeForce GPU. Not because it magically turns an RTX 5070 into a 4090, but because DLSS Frame Generation works reliably, is supported in more games, and includes extras like Multi Frame Generation – not especially useful, but still something AMD lacks entirely. If AMD eventually solves the frame pacing issues across all games, FSR Frame Generation could become a suitable alternative, but right now it’s not close enough.
So unfortunately, FSR Redstone is off to a rocky start. Beyond Frame Generation, we have Radiance Caching (which won’t arrive until 2026), Ray Regeneration in a single game that we’ll examine soon, and FSR Upscaling, which remains a decent feature but unchanged from earlier this year, including no official FSR 4INT8 support.
We like that AMD is investing in new technologies to improve its graphics software stack, it’s sorely needed, but we’re still some distance away from Redstone having a significant impact for gamers.