Has AMD fixed FSR frame generation? That's certainly the question on my mind after the latest suite of FSR releases. The big headline story that captured everyone's attention was the release of FSR 4.1 for RDNA2 graphics cards. But tucked away in the announcement was a brief mention of new enhancements for FSR frame generation with more consistent frame pacing for smoother gameplay. That's [snorts] interesting to say the least.
The biggest issue with AMD's collection of FSR technologies, at least from a functionality standpoint, is broken frame pacing in their frame generation feature. This has been an issue since the very first release of FSR3 technology back in 2023. It remained an issue with the FSR4 Redstone release in 2025, and last I checked, it was still a problem as recently as a few months ago. Broken frame pacing completely undermines the purpose of frame generation technology. The goal of Frame Gen is to improve the smoothness of gameplay. But if generated frames aren't delivered to your display evenly and consistently, the experience won't be smooth. These jitters, stutters, and judders from erratic frame pacing can result in frame generation reducing the smoothness of gameplay or not improving the smoothness compared to disabling the feature. If you don't actually get a boost to smoothness, it's not worth using frame gen at all.
I've occasionally checked in with FSR Frame Gen over the years to see how frame pacing is getting along, and on all previous occasions, I found it to be anywhere from somewhat to significantly worse than DLSS frame generation frame pacing. Nvidia's technology is essentially flawless in this area. It just works and it doesn't require messing around with specific configurations. AMD's solution in the same configuration with the same display hardware is broken.
One of the roadblocks I've come across over the years is a denial that this is a problem. Back at the launch of FSR3, AMD told me this wasn't actually an issue and it was something wrong with my setup. I've been able to replicate the issue on several entirely different PCs and Digital Foundry also showed issues in their coverage. Lots of commenters have claimed that we just aren't testing the feature right, that it requires forcing VSYNC and using specific AMD settings to get it working properly, which isn't really true, and it ignores that you don't need to do any of that when using DLSS frame gen. Ultimately, despite my best attempts to highlight the problem and get it fixed so Radeon gamers could get more out of this feature, it didn't seem like I was really getting anywhere.
So, I was positively surprised when AMD finally acknowledged that frame pacing is a problem with FSR frame generation in this latest FSR announcement. Of course, that acknowledgement comes with a supposed fix for the issue at the same time. So, maybe AMD didn't want to talk about this until they had a fix ready to go, but I'll take it as a win. The question now becomes whether they've actually fixed frame pacing, and I'll get to that after this.
Today's video is brought to you by the AOC Agon Pro AG276QSG2, a brilliant 1440p 360Hz gaming monitor featuring Nvidia G-Sync Pulsar technology. You will not find a better motion experience today. G-Sync Pulsar combines the elite clarity of strobing with variable refresh rates, ensuring you get clear, fast-moving visuals no matter the situation. I tested G-Sync Pulsar earlier this year and was super impressed with how easy it is to use, how flawlessly it works, and how it's able to beat even the fastest OLEDs of today in motion clarity. The AG276QSG2 also includes an adaptive light sensor which can automatically adjust the screen to match your environment, a neat built-in headphone holder, and a fully adjustable esports stand. Gamers that demand a fast display will love G-Sync Pulsar. So, to learn more and grab an AG276QSG2 today, check the links in the description.
Answering the question of whether AMD has fixed frame pacing is more complicated than it appears at the surface. Initially, I thought this would be a simple video dive into a couple of games that I knew were broken previously, see whether they work now, and then call it either way. But there's a lot more going on here than I first realized. The good news is that AMD has fixed frame generation frame pacing in some games. Only some games, you say? Well, that's the complicated part, but basically there are some titles where the latest 26.10.1 driver improves frame pacing on RDNA4 GPUs. And I say driver here because the fix is automatically applied at the driver level. It does not rely on the frame generation override toggle being enabled for specific games. Install the driver and all the work is done immediately.
I'm going to show some slow motion camera footage once again throughout this video, and apologies for the resolution here as the 960 fps footage is not the highest quality, but it is sufficient for analyzing frame pacing. The reason why I'm using a camera to capture this and not just looking at tools like PresentMon with its displayed frame time metrics will become clear later in this testing.
So, here's Crimson Desert running on the RX 9070 XT with frame generation enabled and the latest GPU driver installed. There's a few metrics on the screen here which are sort of visible at this resolution. The big FPS number in the top right is the current in-game frame rate, and directly below that is the average FPS chart. Then we have the game's raw frame time output as a graph. So, this is a game-level measurement of frame times. And at the bottom with the orange line is the displayed frame time, which is supposed to show how long each frame is actually shown on screen. This is still a software measurement of frame times, but it's supposed to measure the actual pace of frame delivery to the display after techniques like frame gen are applied, as opposed to the app frame time which captures different earlier positions in the output chain and may miss the effects of frame gen.
Essentially, displayed time is supposed to be the most realistic and real-world measurement of pacing that PresentMon offers. Additionally, in the top left of the display is the frame rate readout built into this monitor's OSD. This is a monitor-side calculation of frame rate based on what the monitor is receiving. And on this MSI monitor, the readout is updated per frame. All of these data points allow us to validate what is going on. For the best pacing, we want the frame rate on the right to match the monitor's frame rate on the left. This means the game's output and monitor's operation are properly in sync. And of course, we're using variable refresh here. Additionally, we want the frame pacing to be visibly even in the slow motion footage, and we want the real-world gap between frames as measured in this footage to match the readouts on the display. The frame time charts give us additional insights into what is happening. But the ground truth here is what the camera actually captures.
Here in Crimson Desert after the latest driver update, frame pacing is broadly consistent. There's a few hitches here and there as the game itself doesn't always pace frames evenly, but generally speaking, this is quite smooth. The recorded frame rate of the game is matching the monitor's rate. There's no persistent double-tap output like I saw previously. This is a good result.
In comparison, this is what frame pacing in this title used to look like with a previous driver. This is exactly the same GPU and version of the game, but I rolled back to AMD's June driver to reconfirm frame pacing used to be poor and that this was fixed in the latest driver, not because of a game patch. You can see the double-tap effect and the high levels of inconsistencies between the game frame rate and monitor rate.
Another bad example for frame pacing previously was Resident Evil Requiem. This is how the game used to operate with the previous driver: a persistent double-tap output, and the displayed frame time chart is erratic, showing very uneven delivery of frames to the display. This is what the game looks like with the latest driver. The displayed frame time chart is almost flat now. Pacing is obviously more even, and the various frame rate figures are aligning. We're finally getting a smooth gameplay experience here.
I checked out a range of other games just to see how things are working. Here's The Witcher 3 remastered. Perfect pacing here, no issues. Not a game I looked at previously as it wasn't remastered yet, but it's working well today. Assassin's Creed Black Flag Remastered was a title that had issues previously. It works much better after this latest driver update. And while pacing is not completely flawless here, it's much improved with no consistent double-tap.
I also showed issues with Alan Wake 2 previously. These have been resolved now, and frame generation delivers consistent frame pacing on RDNA 4. Lego Batman: Legacy of the Dark Knight has been improved as well. Quite happy with the results here. Nothing more to say really. Gears of War: E-Day is the newest title I looked at, and again, I found no significant issues with frame pacing, only the occasional hitch, which I attribute more to the game engine than to issues with FSR.
However, not every game has been fixed. God of War Ragnarok still pretty erratic, I would say. The displayed frame time chart is jumping all over the place like we saw in Crimson Desert and Resident Evil Requiem prior to the latest driver update. Except this time, God of War is running on the latest driver. So, that's a bit unfortunate.
I saw a similar lack of improvement in Hogwarts Legacy. Frame pacing consistency is poor, especially in this area where I tested. Erratic display time chart, double-tap output, visually the whole works. Borderlands 4 seemed to still struggle as well. Not as erratic as Hogwarts, but still far from the smoothness I was getting in titles like Alan Wake and The Witcher 3. In Borderlands 4, there was often quite a significant gap between the game's frame rate and monitor's rate, and the monitor rate jumped around a lot while moving. I saw similar results in Black Myth: Wukong. That's another title I would say has not been fixed after this driver update.
At this point in the testing, bit of a mixed bag, right? In seven of the titles I looked at, frame generation frame pacing was very smooth and clearly improved in the latest driver update. But then in four titles, the same old issues were cropping up once again: inconsistent frame time delivery to the display, a persistent double-tap issue, and various frame counters jumping around all over the place.
There's a reason why some games are improved and others are not. It's not random. All of the titles that saw fixed frame pacing after the driver update have native FSR4 frame generation game integrations. These titles do not rely on the driver override feature to access the Redstone ML-based version of frame gen. They include it directly into the game and have a separate FSR4 frame gen DLL file. All the titles I tested that directly integrated FSR4 were able to benefit from AMD's fix. And it's good to see that some developers have actually gone back to upgrade the native FSR integration to FSR4, which allows this fix to work. Alan Wake 2 is one of those games.
All of the titles that use FSR 3.1.5 or earlier do not see fixed frame pacing on RDNA4 after this driver update. And it doesn't matter the DLL configuration here. It's more the version of frame gen that's integrated. God of War Ragnarok, for example, has the single combined FSR DLL file, which isn't fixed, while Stalker 2 uses the split configuration with separate DLLs for upscaling and frame gen. Stalker 2 has inconsistent frame pacing despite having a dedicated frame gen DLL, and that's because the version integrated into the game is FSR 3.1.5.
In games like Stalker 2 that have a separate FSR 3.1.5 frame gen DLL, you can fix frame pacing by manually replacing that DLL with the new FSR4 frame gen DLL. This allows AMD's driver fix to work. However, you can't do a DLL swap like this with older titles that use the combined DLL or titles that don't have an FSR DLL at all.
Among the games with native FSR4 integrations, there are a couple of outliers. One is Cyberpunk 2077. At first glance, you'd probably think everything is operating normally here. The monitor's frame rate and the game's frame rate are matched, and the display time chart is flat and consistent. Even just viewing the output, it appears as though frame pacing is even. However, the reported frame rate and the actual frame rate are different. The slow motion footage reveals the interval between frames is approximately 13 milliseconds instead of the 6.5 milliseconds shown in the display time graph, which is indicative of around 77 FPS, not the 150-plus FPS being displayed in the various counters.
What appears to be happening is the game is just not showing the generated frame at all and is repeating the rendered frame twice, which would not be clear from the otherwise normal metrics. For example, if we compare this 140 fps footage from Crimson Desert to this 150 fps footage from Cyberpunk, it's clear the real Cyberpunk frame rate is much lower.
What's unusual about this game is that frame pacing is fixed if you select FSR3 frame generation in the settings instead of FSR4. This title allows you to switch between the two, and for some reason only the FSR3 option has proper frame pacing, but of course the reduced quality of the non-ML-based algorithm. I have no idea why this is the case.
The other outlier is Mafia: The Old Country, which actually does have fixed frame generation thanks to the latest driver update, as confirmed by the slow motion footage. However, what is interesting here is that unlike all the other games assessed, the displayed frame time chart is extremely erratic and very inaccurate. There's just something about this game where the actual frame pacing is good, but it doesn't show up in the software measurement tools.
And this brings up an important point about measuring the performance of FSR frame gen. Tools like FrameView, PresentMon, and its derivatives that show displayed time and ms between display change usually report frame times that align with the actual output of the display, but not always. In games like Cyberpunk, you may find repeated frames instead of a proper output, while in Mafia, the output is smooth when the frame time measurements say that it isn't. When frame generation is enabled, these software tools are simply unreliable, which is why I've stuck to using slow motion footage in this video to verify what is truly happening.
I'm yet to find a software-based measurement tool that actually captures the real-world frame pacing of frame generation accurately and consistently, though these days most are based on PresentMon, which has these inherent issues. The other thing to be aware of is that 1% low data is generally inaccurate when frame gen is enabled. A lot of these titles displayed 1% low information that is far too low considering the actual output performance I was seeing on the display. This is usually because 1% lows are calculated from present frame times instead of displayed frame times. But even if they used displayed times, it would run into the aforementioned issues where that data isn't always accurate. Just something to be aware of when benchmarking frame gen.
So I've pretty much gotten to the bottom of frame pacing on RDNA4 GPUs. If the game has a native FSR4 frame generation implementation, the 26.10.1 driver will largely fix frame pacing in most instances. If the game uses FSR 3.1.5 or older, frame pacing will remain broken. But what about on older architectures? Here's where things get really interesting and a bit confusing.
So firstly, I fired up my Radeon RX 7900 XTX to check out the latest driver on an RDNA3 GPU and found that games like Resident Evil Requiem had good frame generation frame pacing. Cyberpunk 2077 also looked good. Not quite as consistent as I saw from the 9070 XT after the latest driver update, but certainly better than the 9070 XT results using the older driver where frame pacing was erratic.
Initially, I thought that AMD had improved frame pacing on the RDNA3 architecture, but perhaps wasn't able to achieve the same level of pacing as RDNA4. After all, the latest driver supposedly enabled hardware flip metering, a capability of the RDNA4 architecture that isn't found in RDNA 3. Maybe there are general enhancements to frame pacing, and flip metering helps to clean up RDNA4 even further.
Then, as a sanity check, I decided to fire up a few games that still had problems on RDNA4, like God of War Ragnarok. Much to my surprise, this title had much better frame pacing than I was expecting. Clearly better than what I was getting on the 9070 XT. Was this just this specific model of GPU? Well, I put in a 7800 XT and confirmed the same behavior. Decent, though not perfect frame pacing in this title. And here's Hogwarts Legacy, another problematic title that seemingly works quite well on RDNA3 with frame gen enabled in contrast to RDNA4.
This had me quite confused. Had AMD managed to improve frame pacing on RDNA 3 to a greater extent than RDNA4? So, I went down a driver rabbit hole. And to cut a long story short, here's what I found. Basically, the 26.10.1 driver actually does not improve frame pacing on RDNA3 GPUs at all. I saw similar results when rolling back to the June driver. However, frame pacing is improved with the newer drivers compared to the original FSR3 launch driver from 2023.
This is what God of War Ragnarok's frame pacing looks like with that driver, which is pretty similar to what I saw testing FSR3 back in the day in titles like Immortals of Aveum. I didn't test every single driver to find out exactly when AMD implemented these changes, but it seems that at some point in 2024, AMD were able to partially clean up FSR frame generation frame pacing through the driver. It doesn't fully fix the problem like we see from this latest update for RDNA4.
There are still inconsistencies and moments of erratic behavior like I'm showing now in God of War with the latest driver, but this result on RDNA3 is clearly superior to the current output in this game on RDNA4.
In my previous videos on FSR frame generation, I concluded that AMD had not fixed frame pacing with the latest FSR4 Redstone version running on RDNA4 when compared to frame pacing in the original FSR3 release on RDNA3. I hadn't done much testing between the release of the FSR3 and FSR4 versions. I'd occasionally noticed frame pacing issues with FSR3, but I didn't really do a deep dive. I simply assumed that because I was seeing issues on RDNA4's Redstone, that it must be the same issues that were around with FSR3 from years prior.
But what actually seems to have occurred is that AMD were able to somewhat improve frame pacing on RDNA3, only for frame pacing to regress significantly on the RDNA4 architecture. Whatever fixes they implemented previously mustn't have worked on the new GPU architecture. And it's taken this latest driver update to fix the mess of that frame pacing regression.
It's quite an unusual finding. I've tested a bunch of GPUs and different configurations to confirm it, but it does seem like in a stock standard FSR3 title, RDNA3 handles frame pacing better than RDNA4, though neither architecture delivers rock-solid results.
This leads to quite a weird current frame pacing paradigm. In games with native FSR4 integrations, RDNA4 delivers very good frame pacing, near perfect in most situations, and pretty much what you see with DLSS frame generation on GeForce GPUs. FSR4 titles fall back to FSR3 on RDNA3, and while frame pacing here is okay in some situations, it's not as smooth as RDNA4 after the latest driver fixes.
But then in older titles that use FSR 3.1.5 or older for their frame gen, RDNA3 actually outputs better frame generation than RDNA4. The driver fix doesn't work in these games, and the frame pacing regression prior to the driver fix means that RDNA4 delivers a worse output than RDNA 3. It's bizarre, but that appears to be the current situation, as best as I can tell.
So, what to make of it all? Well, it's certainly good that AMD have actually been able to resolve their frame pacing problems via a driver update for RDNA4 owners in titles with native FSR4 integrations. That means a lot of modern games now have proper smooth frame pacing when FSR frame generation is used. And moving forward, I expect this to be the norm as games have been natively integrating FSR4 for a while now. So, at least for new games, it's fixed and we shouldn't have to worry about it anymore.
It's unfortunate that AMD haven't been able to solve frame pacing problems in pre-FSR titles on the RDNA4 architecture. And it's especially frustrating when it seems that RDNA 3 actually has better frame pacing in these situations. It's not clear to me how much more can be done in the drivers to fix frame pacing in these older games given the current fix relies on the newer FSR4 DLL. I hope AMD can figure something out, but FSR 3.1 was released a year before RDNA4. It was possibly designed in a way that can't be properly fixed on the latest architecture.
What we're left with overall is a partial fix for frame gen frame pacing. Fixed in the latest games, and I'm not expecting to encounter issues in new games moving forward, but there are still plenty of titles that suffer from frame pacing issues. The situation on RDNA3 isn't perfect either, and without hardware flip metering, it might remain that way forever.
The last thing to reiterate is that based on what I've seen throughout this testing, there's currently no good software solution to accurately measure frame generation performance. Display frame times get close, but there are still some titles that do weird things, and I wouldn't want to trust what it spits out. That makes validating the actual frame pacing of a game difficult because you basically have to use an external tool to look at what is going on. Hopefully in time some sort of new metric or new form of capturing data will consistently work with frame generation enabled.
Anyway, that's it for this one. If you do want to support Hardware Unboxed and the independent testing that we do, please consider signing up to our Patreon. Links to that is in the description below. If you sign up, you get access to things like our Discord community, monthly live streams, BTS content, and plenty of other good stuff.
That's probably going to do it for this FSR frame generation, frame pacing testing, unless some other, you know, something comes up. Maybe AMD says, 'Hey, we're going to fix this for some of the older titles.' Not 100% sure, but for now, it is good to see some fixes. So, hopefully we don't have to pull out the slow motion camera again for this sort of testing. So, that'll be nice. Anyway, as I said, that's it for this one. Thanks for watching, and I'll catch you in the next one.