AMD really needs FSR 4 to be ?? ???great. Throughout the last two GPU generations, gamers have been abandoning Radeon for GeForce, and a key contributing factor to this is DLSS. Nvidia has continued to improve DLSS upscaling year over year, culminating in their latest release with DLSS 4 which is even better than before.
Meanwhile, AMD has been slipping further behind, especially for gaming at popular resolutions like 1440p, as their FSR 2.2 and FSR 3.1 upscalers just haven't been competitive enough. In most games, Nvidia has been delivering a much higher quality visual experience with upscaling enabled, and Radeon sales have suffered as a result.
Why test at 1440p? Because it is a more common resolution than 4K, and previous versions of FSR struggled at 1440p. To be a truly viable option for gamers, AMD needs to deliver high-quality results at this resolution. It is far more important than 4K – if they get 1440p right, they will almost certainly get 4K right as well.
But that could all change with FSR 4. AMD is finally releasing an upscaler for gamers that's built around an AI model, promising significant improvements to visual quality. Previous versions of FSR used more traditional algorithms, which allowed for broad GPU support, but the quality suffered in comparison to DLSS.
Happy Radeon Day! We're proud to collaborate with @PlayStation on Project Amethyst! ??
– AMD Radeon (@amdradeon) March 6, 2025
FSR 4 is looking fantastic! Excited for the co-development with Sony Interactive Entertainment on the models used for the FSR 4 upscaler.
This is just the beginning. Stay tuned for what's… pic.twitter.com/M8W4dLCkwe
With FSR 4 using AI – AMD says it's a hybrid of a CNN and Transformer model – it's much more likely to be competitive with DLSS. But where exactly does FSR 4 fall? Can it match DLSS 3? Beat DLSS 3? Match DLSS 4? That's what we will be exploring in detail throughout this article and the accompanying video footage.
To assess FSR 4 image quality, we will be analyzing 14 different aspects of upscaling, including texture quality, edge stability, ghosting, disocclusion, foliage, particles, water, and more. The goal is to compare FSR 4 to DLSS, so we are carrying over many of our tests from recent DLSS 4 evaluations. However, FSR 4 game support differs slightly from DLSS, so there are some new titles included, though the areas of focus remain the same.
How Does FSR 4 Work?
Unlike previous FSR versions that were designed to be broadly compatible, FSR 4 only works on RDNA 4 graphics cards like the Radeon RX 9070 XT. This is because it uses FP8 processing, which isn't accelerated on previous RDNA GPUs. As a result, FSR 4 cannot be used on RDNA 3 or older GPUs, and likely never will be, even though AMD has left the door open for potential support.
Accessing FSR 4 is very similar to Nvidia's latest implementation of DLSS 4. Some games have native integrations – simply enable FSR 4 in the game, and you're good to go. For other titles, you can upgrade them from FSR 3.1 to FSR 4 using AMD's driver, similar to Nvidia's DLSS override feature.
However, there are a couple of important things to note. First, a game must have implemented at least FSR 3.1 to be upgrade-compatible, as that was the first release to use a DLL. Games using older versions of FSR, like 2.2 or 3.0, cannot be upgraded, which is disappointing since those versions make up the majority of FSR-supported titles. Second, AMD uses a game whitelist similar to Nvidia, so only approved FSR 3.1 titles have the FSR 4 upgrade toggle.
At present, most FSR 4-supported games are getting FSR 4 via the driver upgrade toggle rather than native game integration. However, during the review period, some FSR 3.1 titles were not supported, such as Stalker 2 and Indiana Jones and the Great Circle. Presumably, those titles will need to be whitelisted, making future driver updates crucial.
We are kicking off the analysis of FSR 4 with a look at 1440p. Why 1440p? Because it is a more common resolution than 4K, and previous versions of FSR struggled at 1440p. To be a truly viable option for gamers, AMD needs to deliver high-quality results at this resolution. It is far more important than 4K – if they get 1440p right, they will almost certainly get 4K right as well. This is a stress test, and we are holding nothing back.
All FSR 4 examples were captured at 1440p using the Radeon RX 9070 XT, while all other examples were captured using the GeForce RTX 5090. A selection of games was used, with settings such as motion blur, grain, vignette, and chromatic aberration disabled.
Sharpness was set to 0 for upscaling unless otherwise specified. All DLSS 3 examples were upgraded to DLSS 3.8.10, except for Hunt: Showdown, which remains on DLSS 3.7 as it cannot be upgraded. All DLSS 4 examples utilize the latest model via Nvidia's driver override.
Let's get into it.
FSR 3 vs FSR 4
FSR 4 is a significant improvement over FSR 3.1 in almost every way. FSR 4 is much less blurry in motion and far more stable. In areas with fine detail, FSR 4 is a night-and-day difference in reducing sizzling and grain, particularly in elements like hair and grass.
For a while, we thought we might have to censor FSR 3.1 images because YouTube doesn't allow offensive content on the platform, but after checking the guidelines, we think it's okay – even if it has the potential to scorch your eyeballs. Just cover your eyes if it gets too ugly.
For a better representation of image quality comparisons, check out the HUB video below:
The difference is so massive and so obvious while gaming that it barely seems like these two technologies belong to the same family or even deserve the same name. This is almost like the leap from DLSS 1 to DLSS 2.
This improvement is evident in every example using the Quality mode. If we switch to the Performance mode, it's actually surprising that AMD allowed people to use the FSR 3.1 Performance mode at 1440p because it produces terrible image quality – really, really awful.
FSR 4 is actually somewhat usable at 1440p in Performance mode. Yes, there are some issues in motion, but not the kind that blur and sizzle the entire screen. The difference is so significant that it's hard to describe – if the FSR 4 Performance mode represents true 1440p, then the FSR 3.1 image in comparison is, what, like 360p at best?
We were blown away by some of these comparisons because we weren't expecting FSR 4 to improve this much in Performance mode at 1440p. FSR 3.1 was so far behind DLSS in this configuration that we expected an improvement but still anticipated some level of disappointment. Gone are the days of FSR being completely useless at upscaling from a low render resolution.
Now, the real testfor FSR 4 is how it compares to DLSS 3 and DLSS 4. Beating FSR 3.1 is nice, but that bar was relatively low.
The real challenge is competing with the industry leader in upscaling, so for the rest of this article, we will primarily compare FSR 4 with DLSS. How do these technologies compare in terms of blur and texture quality?
Textures and Blur
FSR 4 holds up really well in this regard. Previously, we found that DLSS 4 had practically eliminated TAA blur in motion, even down to the Performance mode at 1440p. This meant that texture quality was enhanced and the overall image appeared sharper relative to DLSS 3 and even native rendering. The same applies when comparing FSR 4 to DLSS 3 using the same modes – in this case, Quality vs. Quality.
For a better representation of image quality comparisons, check out the HUB video below:
In this Spider-Man 2 example, we are rapidly switching between a still shot while stationary and a still shot captured mid-motion. The DLSS 4 version preserves almost the same level of detail and sharpness whether stationary or in motion. DLSS 3 appears quite blurry in motion compared to the stationary image. FSR 4 sits somewhere between the two, with significantly better clarity in motion than DLSS 3, but not quite reaching the level of DLSS 4.
This trend was observed in other games we tested, though results varied depending on the title and scene. For example, in Hunt: Showdown, FSR 4 was much closer to DLSS 3 than DLSS 4 in terms of clarity. However, in The Last of Us Part I, the reverse was true, and FSR 4 was much closer to DLSS 4. In all of these examples, in-game sharpness was set to 0, ensuring no additional sharpening was applied after the upscale pass.
It also doesn't seem like FSR 4 is simply sharpening a blurry image; rather, it appears to be handling fine texture details in motion more effectively. There are some aspects of the DLSS 3 image that cannot be sharpened to match the clarity of FSR 4, which generally gives AMD's upscaler the edge. This is great news since image crispness is a major selling point for DLSS 4, and in many cases, FSR 4 is competitive in this aspect.
The performance of these upscalers across various quality settings remains relatively consistent. This means that the clarity of the Quality mode and the clarity of the Performance mode follow similar trends.
In some instances, DLSS 4's Performance mode is sharper and better than FSR 4's Quality mode, though this doesn't necessarily hold true for other aspects of image quality. There are also situations where distinguishing which upscaler provides better clarity requires zooming in, which is a great result for FSR 4. In most cases, we preferred its sharpness over DLSS 3.
Edge Stability
Next up is edge stability. You might have noticed in some of the previous shots that FSR 4 can struggle with maintaining stable edges in motion, particularly with characters. However, this issue was primarily observed in the two Insomniac games we used for testing – Ratchet & Clank: Rift Apart and Marvel's Spider-Man 2 – both of which use the same game engine.
For some reason, in these specific games, FSR 4 is more prone to aliased character edges compared to DLSS 4 and DLSS 3, which can look unappealing in certain scenes.
For a better representation of image quality comparisons, check out the HUB video below:
However, when switching to similar third-person titles like Horizon Zero Dawn Remastered and The Last of Us Part I, this aliased edge issue was not present, and the stability of character edges in motion appeared fine. It's unclear why this happens, but in a worst-case scenario, FSR 4 does not preserve edge stability in motion as well as DLSS 3.
Outside of these cases, FSR 4 delivers mixed results. While it represents a huge step forward compared to FSR 3, in the best examples, it only matches DLSS 3 in stability – while DLSS 4 remains far ahead in preserving stable edges and details between frames.
For example, in Ratchet & Clank, both FSR 4 and DLSS 3 performed similarly in maintaining stair edges as the player approached them. A similar trend was observed in Spider-Man 2. In Hunt: Showdown, aliasing and shimmering levels were comparable when looking at FSR 4 Quality versus DLSS 3 Quality, with DLSS 4 clearly superior in this regard.
In other situations, FSR 4 is not as stable as DLSS 3 when comparing the same quality settings. For instance, in The Last of Us Part I, FSR 4 struggled in slow pan shots, creating instabilities on bricks and window edges – artifacts that were absent in the DLSS 3 image and could be somewhat distracting.
Strangely, in The Last of Us Part I, FSR 4 Balanced mode actually outperformed DLSS 3 Balanced in some cases, with little difference in stability between the two FSR 4 modes. At times, FSR 4 Balanced even appeared more stable than FSR 4 Quality, which is unexpected. FSR 4 winning in a Balanced vs. Balanced comparison was quite common, meaning that at worst, FSR 4 is only half or even less than half a quality tier behind DLSS 3.
The challenge AMD faces is that DLSS 4 Performance mode is often more stable than DLSS 3 Quality mode, which means DLSS 4 Performance also surpasses FSR 4 Quality. It's possible that AMD prioritized reducing blur over improving stability, as FSR 4 is generally the sharper but less stable image compared to DLSS 3. Meanwhile, DLSS 4 delivers the best of both worlds. This is one of the biggest areas AMD needs to focus on when improving future versions of FSR 4.
Ghosting
When it comes to ghosting, all upscalers are prone to it in the worst cases. Fortunately, across most of the examples we analyzed, FSR 4 is not particularly susceptible to ghosting and does not introduce ghosting in areas where DLSS does not.
Most of the time, performance is comparable, so for example, when examining particle effects in Ratchet & Clank, both DLSS 3 and FSR 4 exhibit minor ghosting, just at different times. However, DLSS 4 is relatively more resilient in this scenario.
There's another example from Spider-Man 2, where each algorithm is triggered by slightly different conditions, but no technique is completely immune. Unfortunately, we couldn't put FSR 4 through the Forza Motorsport test since FSR 4 is not supported in that title. However, in the games we did test, FSR 4 is unlikely to cause significant ghosting issues.
Disocclusion
A major win for FSR 4 comes in the area of disocclusion. Among the three main techniques examined here, FSR 4 generally provides the best image quality, or at the very least, matches DLSS 3, which was previously the leader.
As we found in the DLSS 4 analysis, DLSS 4 prioritizes a sharper image, often resulting in worse disocclusion than DLSS 3, leading to more sizzling around characters in third-person games. In titles such as The Last of Us Part I, this trend holds when comparing DLSS 4 to FSR 4 – AMD's technology comes out ahead with fewer noticeable artifacts at the same Quality mode.
For a better representation of image quality comparisons, check out the HUB video below:
It is harder to distinguish differences in games like Horizon Zero Dawn, but upon closer inspection, there is slightly less fizz around Aloy when running in the FSR 4 images compared to DLSS 3, and especially DLSS 4.
In Hunt: Showdown, the DLSS 4 image appears slightly smeared around the gun in motion, while FSR 4 remains relatively clean with only a few disocclusion artifacts, with DLSS 3 performing similarly to FSR 4.
In many situations, disocclusion artifacts in DLSS 4 are not particularly noticeable, as they typically disappear within a few frames, making them difficult to detect during real-time gameplay. However, the most impressive aspect here is that AMD has gone from having the worst disocclusion handling, with frequent sizzling around characters, to essentially having the best implementation at 1440p.
Hair
Hair quality is very similar across all three upscalers. When comparing images in Quality mode, we did not find many substantial differences in motion, even in challenging scenarios such as Aloy's hair billowing as she runs.
The detail, stability, and clarity of the hair are virtually identical between FSR 4 and DLSS, meaning the ugly grain seen constantly with FSR 3 is no longer an issue. This element is now being upscaled effectively.
This applies across various games, including fur rendering in Ratchet & Clank. The only major difference is that the edge of the fur appears aliased using FSR 4. However, this issue seems to be related to the aliasing problems observed in Insomniac games, as discussed earlier.
This type of artifact is not present in a title like The Last of Us Part I. Speaking of The Last of Us, it was reassuring to see decent beard stability, as well as slightly improved hair rendering in one specific example when comparing FSR 4 to DLSS.
It is also not a case where DLSS outperforms FSR at lower quality settings. When lowering DLSS to Balanced while keeping FSR 4 in Quality mode, FSR typically appeared slightly better in this comparison. This represents a strong result for AMD's upscaler in this area.
Particles
Particle quality presents mixed results. When observing confetti in Ratchet & Clank, there is a significant difference in how FSR 4 renders it compared to DLSS 3 and 4. With FSR 4, the confetti appears much sharper and clearer, but this also exposes more aliasing on the edges, which is less noticeable in the softer DLSS presentation.
As a result, the confetti can flicker slightly in an unzoomed image, whereas DLSS 4, in particular, produces a much smoother output. However, FSR 4 demonstrates noticeably less ghosting in this area compared to DLSS 3, making it the superior option in some respects. That said, overall, DLSS 4 looks the best, with FSR 4 Quality only really being comparable to DLSS 4 Performance.
For a better representation of image quality comparisons, check out the HUB video below:
The results were different in The Last of Us Part I when examining dense spore clouds. Here, FSR 4 looked particularly strong, especially in comparison to DLSS 3, which suffered significantly from ghosting. Compared to DLSS 4, FSR 4 actually reconstructed more spore particles, though DLSS 4 handled minor ghosting and shimmering more effectively.
Based on these examples, FSR 4 particle quality should generally be superior to DLSS 3 in most scenarios and potentially approach DLSS 4 levels.
Transparency
We were very impressed with the quality of transparencies in FSR 4. This was a major weakness in FSR 3 and 2.2, where transparencies appeared garbled and looked terrible.
With FSR 4, transparencies are much clearer, more stable in motion, and have better detail, eliminating the poor quality seen in previous versions. In fact, FSR 4 is on par with DLSS 3 and 4, with DLSS 4 perhaps having a slight edge.
Not only are transparencies well-handled while standing still – something that was often an issue with FSR 3 – but fire also benefits significantly from superior reconstruction. Grainy, pixelated fire is much less common in FSR 4, resulting in image quality similar to DLSS.
In Ratchet & Clank, the rotating hologram is clearer and resolves detail better, though DLSS 4 is slightly less aliased for some of the fine lines. The transparent logo on top of the store character now looks like any other element in the game.
This also improves scenarios where characters or objects become transparent as the player moves through the game world. For example, in Spider-Man 2, when turning the camera close to a wall, Spidey briefly becomes transparent. This transparency looks very similar between DLSS and FSR 4, with DLSS 4 exhibiting slightly better edge quality, but FSR 4 still performing well.
Comparing this to FSR 3 highlights just how much of an improvement has been made, even in Balanced mode. This is a huge win for games featuring many transparent objects that require upscaling.
Fine Details
When it comes to fine detail reconstruction, FSR 4 generally does not perform as well as DLSS 3. While the two technologies can appear similar when stationary, in motion, DLSS 3 is typically more stable and better at preserving detail in fine wires and lines on distant objects.
This is especially noticeable in scenes from The Last of Us Part I, where at 1440p, FSR 4 is the least stable presentation among the options tested, while DLSS 4 is significantly more stable.
For a better representation of image quality comparisons, check out the HUB video below:
This issue is most apparent on very fine pixel or near-pixel level wires, where FSR 4 can become aliased in motion relative to DLSS. While the DLSS 3 image is slightly blurrier, it maintains better stability. However, FSR 4 is not far behind, and in examples like Hunt: Showdown, both DLSS 3 and FSR 4 exhibit similar stability issues for fine elements.
DLSS 4 is clearly the best option for fine detail stability, meaning a lower DLSS 4 mode can often match or surpass FSR 4. In fact, there were cases where the Performance mode in DLSS 4 appeared superior to FSR 4 in Quality mode. This aligns with earlier findings on general edge stability – FSR 4 is not quite at the same level when rendering at 1440p, and this remains one of its primary weaknesses.
Trees
Tree quality is another area where FSR 4 delivers mixed results. The simplest way to describe its performance relative to DLSS 3 is that it is sharper but less stable. This creates a trade-off – depending on individual preference, some may favor the sharper FSR 4 image, while others may prefer the smoother DLSS presentation.
Several factors, such as motion speed and tree density, influence these results. For instance, in The Last of Us Part I and Horizon Zero Dawn, where a single tree is set against a relatively clean background, FSR 4's stability issues are less apparent, while its clarity advantages are more noticeable.
However, in scenes where trees overlap, FSR 4 struggles more compared to DLSS 3, as the blurrier DLSS image helps mask potential instabilities. DLSS 4 offers the best of both worlds, often delivering similar or better clarity than FSR 4, with enhanced leaf detail and improved stability.
That said, DLSS 3 does not always outperform FSR 4. In this Hunt: Showdown example, DLSS 3 maintains better stability for fine tree branch details, but the foliage itself is more stable in the FSR 4 image.
Additionally, as the quality modes are lowered, FSR 4 gains ground relative to DLSS 3. In Balanced mode, the two techniques appear similar, while in Performance mode, FSR 4 has the edge. It is only in the highest quality setting that DLSS 3 holds a noticeable advantage.
Overall, tree artifacts in FSR 4 are not particularly distracting, unlike in FSR 3, where trees often sizzled in motion and looked terrible. With FSR 4, there is a good chance that at 1440p, these artifacts will go unnoticed, making them less of a concern compared to other stability issues discussed earlier.
Grass
Grass quality in FSR 4 is very impressive, especially considering this was a major weakness in previous versions of FSR. Gone are the days of grass appearing extremely grainy and ugly in motion. At 1440p using the Quality mode, FSR 4 renders grass with detail, general stability in motion, and no grain or sizzling issues. In games like Horizon Zero Dawn, it looks fantastic, surpassing DLSS 3 in clarity and motion stability, and almost reaching the level of DLSS 4.
Other titles also showed favorable results for FSR 4, such as The Last of Us Part I. While trees can be slightly less stable in FSR 4 compared to DLSS 3, the opposite is true for grass quality, which is generally better in FSR 4. DLSS 4 holds the edge in overall stability, but FSR 4 remains highly usable. In fact, the largest gap between FSR 4 and DLSS 3, favoring FSR, was observed when viewing the very thin strands of grass in Hunt: Showdown. In this case, DLSS 3 Quality mode exhibited some graininess, while FSR 4 looked fantastic by comparison.
FSR 4 did struggle with the unusual purple grass and shrubs in Ratchet & Clank, which were less stable in the FSR 4 image compared to DLSS 3. However, in most other examples, FSR 4 delivered excellent grass quality, often matching either DLSS 4 Quality or Balanced modes.
Fences
Fence quality is another area with mixed results. For fences with thicker details and repeating patterns, FSR 4 generally delivers image quality at or slightly above DLSS 3 levels. It is much less likely to introduce ugly pattern artifacts compared to FSR 3, though it still does not reach the level of stability or artifact-free rendering seen in DLSS 4. The fact that FSR 4 offers a similar or better experience than DLSS 3 is a strong result.
For mesh-style fences with thinner details, FSR 4 struggles more. In this Spider-Man 2 example, there is noticeable sizzling and aliasing in the FSR 4 image compared to DLSS 3 in Quality mode, particularly when multiple meshes overlap.
While DLSS 3 can appear slightly soft when upscaling these details, it produces less shimmering than FSR 4, with DLSS 4 clearly performing best. However, in some cases, such as this example in Hunt: Showdown, FSR 4 holds up relatively well against DLSS 4.
A particularly strange finding was that in some fence scenarios, FSR 4 Balanced mode appeared more stable than FSR 4 Quality mode at 1440p, such as in the Spider-Man 2 overlapping meshes example.
After triple-checking, we consistently found the Balanced mode to provide better visual stability. If instability becomes an issue, it may be worth experimenting with Balanced mode instead of Quality mode.
Cloth
Cloth quality in FSR 4 is generally strong when applied to character clothing and flags flying in the wind. The differences between FSR 4 and DLSS are minimal, making this a solid result for third-person games, particularly those featuring characters with capes.
For a better representation of image quality comparisons, check out the HUB video below:
Regarding moiré effects, there is another odd situation. In a comparison of FSR 4 Quality versus DLSS 3 Quality, FSR 4 is more prone to displaying moiré artifacts, as seen in this Ratchet & Clank example. However, when comparing FSR 4 Balanced mode to DLSS 3 Quality mode, the Balanced mode actually reduces moiré artifacts compared to Quality at 1440p.
While it is still not as good as DLSS 3, the improvement is noticeable despite the lower render resolution. AMD may need to refine the Quality mode further, as there should be no situations where it performs worse than Balanced mode, yet we have already encountered several such cases.
Water
Water quality is very similar across all three technologies. Previously, FSR 3 had significant issues with aliasing on the surface of water, particularly in fast-moving streams with ripples. With FSR 4, surface quality is now comparable to DLSS 3, and in the worst cases, it is only slightly behind DLSS 4. This should no longer be a major concern when using FSR upscaling.
Rain
We were pleased to see that rain quality has been significantly improved in FSR 4. In our punishing snow test – and we are reliably informed that snow is just a magical form of rain – FSR 4 closely matches DLSS in overall quality.
Previously, with FSR 3, rain effects appeared as a blurry, smeared mess, but now FSR 4 provides a much clearer and more usable image, even when comparing Performance mode across upscalers.
In fact, FSR 4 appears slightly better than DLSS 3 in some cases. Snow particles are less blurry, and background disocclusion is handled with less grain, putting FSR 4 very close to DLSS 4 in terms of overall quality.
Performance Benchmarks
Next, we analyze performance. The benchmarking process is straightforward: we tested the Radeon RX 9070 XT using FSR 3 and FSR 4, and the GeForce RTX 5070 Ti using DLSS 3 and DLSS 4.
These cards are closely matched in overall performance, allowing for a direct comparison of the performance uplift provided by FSR relative to native rendering and DLSS relative to native rendering. Since the baseline performance of each GPU is similar, these results should be quite comparable.
We start with Ratchet & Clank: Rift Apart at max settings. As observed previously, DLSS 4 runs slightly slower than DLSS 3 in this title, and we see a similar trend with FSR 4 versus FSR 3. However, the overall performance uplift from each upscaler is very similar. FSR 4 on the 9070 XT provided a 31% increase in frame rate compared to native rendering using Quality mode, while DLSS 4 on the 5070 Ti also delivered a 31% improvement. Balanced mode results were closely matched, while FSR 4 had a slight advantage in Performance mode.
In The Last of Us Part I, we recorded a 22% performance improvement using FSR 4 Quality mode compared to native TAA at 1440p on the 9070 XT. On the 5070 Ti, we observed a 24% improvement, which is very similar. At each quality level, DLSS 4 provided a slightly higher performance uplift, but the difference was minimal.
Horizon Zero Dawn Remastered already runs well on these GPUs at 1440p, so upscaling has a smaller impact. DLSS 4 provided a 16% uplift in Quality mode and a 24% uplift in Performance mode. FSR 4 showed a 14% improvement in Quality mode and a 29% improvement in Performance mode.
Results in Spider-Man 2 were similar. Here, DLSS 4 was 24% faster in Quality mode compared to native, 29% faster in Balanced mode, and 34% faster in Performance mode. FSR 4 was 22% faster in Quality mode, 26% faster in Balanced mode, and 31% faster in Performance mode. While slightly slower than DLSS, the difference was not substantial.
Looking at the geometric mean across this four-game sample, FSR 4 and DLSS 4 provide nearly identical performance boosts at 1440p on similar class hardware.
Only a few percentage points separate the two technologies. In general, we see a little over a 20% boost in Quality mode and around a 30% boost in Balanced mode. Of course, upscaling has a greater impact when the base frame rate is lower, but in these games, native rendering was typically around 100 FPS or higher, reducing the overall impact of upscaling on performance.
What We Learned
When assessing all aspects of FSR 4 upscaling at 1440p, we believe the technology comes out looking quite strong. This represents a massive step forward for AMD's upscaler at this resolution. The differences between FSR 3 and FSR 4 are enormous, providing owners of new Radeon RDNA 4 graphics cards with a significantly improved experience in supported titles.
FSR 3 was mostly unusable at 1440p and not very competitive with DLSS 3, let alone DLSS 4. It suffered from severe artifacts in motion, excessive grain and sizzling, significant blur, and an inability to handle low render resolutions. The signature "fizzled" look of FSR 3 at 1440p has been eliminated in FSR 4, making the technology not just usable, but genuinely effective without major compromises to image quality. This clears the first hurdle FSR 4 needed to overcome: it is not just barely usable but actually performs well.
What's most impressive is that FSR 4 image quality generally falls somewhere between DLSS 3 and DLSS 4. Since many gamers consider DLSS 3 good enough for 1440p gaming, matching or even exceeding that level of quality is a significant win for FSR 4. In most cases, FSR 4 should meet the expectations for image quality, delivering a pleasant experience. This has substantially narrowed the upscaling gap between AMD and Nvidia at 1440p.
What's most impressive is that FSR 4 image quality generally falls somewhere between DLSS 3 and DLSS 4.
FSR 4's biggest advantage over DLSS 3 lies in texture quality and blur reduction. AMD has effectively eliminated TAA blur, something DLSS 3 does not fully achieve. In direct comparisons, this is the most noticeable improvement – FSR 4 is simply less blurry. Additionally, FSR 4 delivers enhancements in disocclusion, grass quality, particles, and rain, often appearing less grainy than DLSS 3 in motion.
That's not to say FSR 4 is not universally superior though. Its biggest weakness is stability, which affects elements like edges, fine details, trees, and fences in motion. In the best cases, FSR 4 can be slightly more stable than DLSS 3 at 1440p, but there are many situations where DLSS 3 holds the advantage.
This creates a trade-off: FSR 4 offers better clarity but reduced stability. AMD needs to refine this area, particularly in cases where FSR 4 Balanced mode is more stable than FSR 4 Quality mode – this inconsistency needs to be addressed. Fixing this issue could further close the gap to DLSS 3. Even in instances where FSR 4 is less stable, it remains usable and is not far behind.
As we mentioned just a minute ago, overall FSR 4 image quality sits between DLSS 3 and DLSS 4, meaning DLSS 4 is still the best option for 1440p gaming. DLSS 4 essentially combines the best aspects of DLSS 3 and FSR 4: it is as clear and blur-free as FSR 4, if not clearer, while also maintaining excellent stability.
In certain situations, FSR 4 comes close to DLSS 4 – such as with transparencies, rain, and hair – but overall, DLSS 4 retains an edge in image quality. The only clear advantage FSR 4 has over DLSS 4 is in disocclusion.
This puts AMD in an interesting position. While DLSS 4 is better and delivers the highest image quality for 1440p gamers, is it worth choosing an Nvidia GPU solely for access to DLSS 4 instead of FSR 4? If this were a comparison between DLSS 4 and FSR 3, or even DLSS 3 and FSR 3, the answer would be a definitive yes – buying a GeForce GPU instead of a Radeon, even at a premium, would be justified.
But now that FSR 4 is highly usable and effective in most scenarios, the decision becomes less clear. If two hypothetical GPUs offer the same performance at the same price – one with DLSS 4 and one with FSR 4 – the GeForce GPU would still be the preferred choice due to DLSS 4's superior upscaling. However, if the Radeon GPU is cheaper, the decision becomes more difficult.
Would it be worth paying 10% more for DLSS 4? 15%? 20%? Previously, a 20% premium for DLSS was justifiable, but now that FSR 4 has significantly improved, that margin seems less reasonable. Right now, we would be satisfied playing games with FSR 4 at 1440p, but would also be willing to pay a small premium for DLSS 4. Beyond a certain price difference, however, it becomes harder to justify spending more on a GeForce GPU unless it offers additional compelling features beyond upscaling.
The next big problem for AMD, the spanner in the works, is game support. The decision to pay extra for DLSS 4 is contingent on games also supporting FSR 4. If that's not the case, if we're instead talking about DLSS 4 versus FSR 3 or FSR 2.2, then that value comparison goes out the window, making a GeForce GPU the obvious choice. In that situation, paying a 20% premium for a GeForce GPU – or even more – becomes justifiable.
The reality is right now Nvidia GPUs offer an acceptable level of upscaling in far more games than Radeon GPUs due to stronger DLSS support. While this may improve as more developers integrate FSR 4, AMD remains far behind Nvidia in game support.
This is largely because Nvidia laid the groundwork years ago with DLSS, which uses a DLL that can easily be upgraded to DLSS 4 via first-party or third-party tools. As a result, most modern titles offer, at minimum, DLSS 2 or DLSS 3, with many now supporting DLSS 4. In contrast, while some games can be upgraded to FSR 4 – and it is great to see a reasonable selection of titles at launch – many remain stuck with FSR 2.2, which significantly impacts real-world game support in Nvidia's favor.
Looking ahead, we hope to see the gap in game support and image quality continue to close. FSR 4 provides a solid foundation for this progress, but AMD still has work to do. The company needs to convince developers to update existing games to FSR 4, including major titles like Cyberpunk 2077, while ensuring future titles include FSR 4 support at launch. At the same time, AMD must continue refining its AI upscaling algorithm to improve image quality in edge cases.
Until that happens, AMD remains in an early adoption phase with FSR 4. Yes, the upscaling gap has narrowed, but it may not be closed in the games you want to play. So you'll need to factor that in to your value calculations when tossing up between GeForce and Radeon. If you strike gold with an FSR 4 game you'll be in for a pretty great experience, and we expect Radeon owners to benefit from increasing game support over the next few years.
Last but not least, based on our testing, our recommendation on quality settings at 1440p, we generally suggest using Quality mode. However, this varies by game. In most titles, Quality mode provides the best image quality, with noticeable compromises when lowering settings. However, in some games, Balanced mode can actually deliver a more stable image. If stability issues arise in Quality mode, switching to Balanced may improve the experience. Otherwise, Quality mode remains the preferred setting for most cases.