On October 20, 2026, AMD expanded the list of devices that support FSR Upscaling 4.1, its machine-learning technology that reconstructs game images at higher resolution, to include Radeon RX 6000 series GPUs and some Ryzen-powered PCs. AI-based upscaling that had mainly been offered on the Radeon RX 9000 series is now officially available on older GPUs and integrated graphics. However, using the new FSR will not necessarily make games faster. Machine-learning image reconstruction can require more computation than conventional methods, so frame rates may drop at the same quality settings. In judging how practical it is on older GPUs, what matters is not only how much image quality improves, but also whether quality holds up when the internal resolution is lowered.
About 13% performance drop on the RX 6900 XT: a trade-off for better image quality
In a quick test by Tom's Hardware, running Cyberpunk 2077 on a Radeon RX 6900 XT at 4K with FSR in Balanced mode, performance dropped by about 13% when FSR 4 was used. The outlet gave this figure while comparing against FSR 3, and said that although the frame rate fell, the image was sharper, particularly in moving scenes, and rendering was more stable.
However, the test does not list specific fps figures before and after the change, and it does not state exactly which FSR version was used for comparison. Conditions such as CPU, ray-tracing settings and whether frame generation was enabled are also unknown. The roughly 13% drop should therefore be treated as the result of a quick test on a specific GPU and game, and it does not necessarily apply to the whole RX 6000 series or to handheld gaming PCs.
It is not technically odd for a new FSR to lower the frame rate. Upscaling reduces rendering load by drawing the game at a resolution lower than the display resolution and then reconstructing the image at a higher one. According to GPUOpen's developer documentation, FSR 4 analyzes temporal and spatial information across multiple frames and uses machine learning to reconstruct image detail. The final frame rate depends on both the load saved by rendering at a lower resolution and the extra load added by reconstructing the image.
For example, if you stay in the same Balanced mode and change only the FSR generation, the number of pixels the game renders internally basically stays the same. If the new reconstruction process is more computationally demanding, the frame rate can therefore fall even as image quality improves. Note that comparing against native-resolution rendering and comparing a switch from existing FSR to the new FSR measure different performance gaps.
According to AMD's official announcement, to support the RX 6000 series it optimized the INT8 model developed for the RX 7000 series. INT8 is a computation format that uses 8-bit integers and is used to reduce the computational load of machine-learning models. AMD says that by adjusting the model and processing so they can run on existing GPUs, it delivers image-quality improvements over FSR 3.1. How much image quality actually improves, and what the impact on performance is, must be evaluated separately, weighing AMD's explanation against measured results in individual games.
If internal resolution can be lowered, better quality may translate into better performance
The value of FSR 4.1 cannot be judged from comparisons at the same quality settings alone. If the new machine-learning model can reconstruct high-quality images from less pixel information than before, it may be possible to change the upscaling setting and cut the rendering load further. For example, in a game that required Balanced mode with conventional FSR, FSR 4.1 might maintain sufficient quality in Performance mode. In that case, even if the computational load of image reconstruction rises, the load saved by lowering the internal resolution could offset it.
According to GPUOpen's "Scaling modes", FSR's standard scaling factors are 1.7x per axis for Balanced mode and 2.0x for Performance mode. Based on these standard values as of October 21, 2026, and fixing the output resolution, the number of pixels rendered internally is as follows.
| Mode | Scaling factor per axis | Internal render pixels as share of display pixels |
|---|---|---|
| Balanced | 1.7x | About 34.6% |
| Performance | 2.0x | 25% |
The share of internally rendered pixels is found by dividing 100% by the square of the scaling factor. Switching from Balanced to Performance gives a reduction of [1 − (1.7 ÷ 2.0)²] × 100 = 27.75%, meaning the number of internally rendered pixels falls by about 27.8%. This is a theoretical value based on the standard factors and does not account for factors such as a game's dynamic resolution or rounding of render sizes to integers.
That said, a roughly 27.8% reduction in rendered pixels does not mean the frame rate rises by the same proportion. CPU-side processing and the various processes that run at display resolution do not shrink at the same rate, and the upscaling itself also takes computation time. GPUOpen likewise explains that lowering the input resolution speeds up reconstruction somewhat, but that its load depends mainly on the output resolution.
Still, if FSR 4.1 can maintain sufficient image quality at a lower internal resolution, users get a wider range of settings to choose from. Tom's Hardware also points out that changing the upscaling mode may make up for the performance loss from the new reconstruction process. However, this describes a possibility of improving the balance between image quality and speed; it has not been demonstrated that equivalent quality can be maintained, or that this test offset the performance loss.
In an actual comparison, it is best first to measure the image-quality and frame-rate differences between old and new FSR at the same upscaling mode, then check what changes when the internal resolution is lowered. This separates the added load of reconstruction from the load saved by rendering at a lower resolution. Comparing not only still-image sharpness but also flicker and edge stability when the camera moves will make it easier to judge the image-quality benefits of the new FSR.
Lightweight models for Ryzen handhelds; Xbox support coming soon
For the Ryzen AI 9 series and some Ryzen Z2 series chips, AMD has prepared a more efficient FSR 4.1 model, separate from the one for the RX 6000 series. It is designed to match the processing power of the target hardware, aiming to deliver practical performance on handheld gaming PCs and ordinary laptops. By applying optimizations suited to the performance characteristics of each APU, which integrates a CPU and GPU on one chip, AMD is increasing the number of devices that can use machine-learning-based upscaling.
For handheld gaming PCs, the supported chips are the Ryzen AI Z2 Extreme, Ryzen Z2 Extreme and Ryzen Z2 A. The Ryzen AI Max series has also been added as officially supported. However, AMD has not said whether the same lightweight model is used for Ryzen AI Max. Even under the common FSR 4.1 name, the same machine-learning model does not necessarily run on every GPU or APU, and processing time may differ. The roughly 13% performance drop reported on the RX 6900 XT therefore cannot be used as-is to predict performance on handheld gaming PCs.
Driver availability also needs attention. The release notes for AMD Software: Adrenalin Edition 26.10.1 list support for the Radeon RX 6000 series, Ryzen AI Max 300/400 series and Ryzen AI 9 300/400 series as new features. However, the driver distributed on AMD.com does not include handheld gaming devices among its supported products and directs users to check the dedicated drivers provided by device makers. Even though the relevant Z2 processors were included in the announcement, the feature may not be immediately available on the handheld you actually use, so you need to check each manufacturer's update status.
AMD has also said it will support the Xbox Series X|S in the future. As of October 20, 2026, however, it is described only as "coming soon," and neither a specific release date nor supported games have been announced. The release of a PC driver is a separate matter from the feature becoming available on Xbox. The actual effect on image quality and performance on the console will need to be evaluated after it is implemented in supported games.
Games that can use FSR 4.1, and limits on supported features
To use the new FSR 4.1, you must also check whether the game supports it. AMD's Japanese official FAQ describes two routes for enabling machine-learning-based FSR Upscaling. One is for games with the machine-learning version of FSR built in directly, which you enable from the in-game settings menu. The other is for games that support FSR 3.1 or later, where you replace the upscaling DLL through AMD Software to switch to the machine-learning version.
In the latter case, with the game closed, open the "Graphics" settings under "Gaming" in AMD Software and enable FSR Upscaling. Then enable FSR 3.1 in the game as well and select the desired quality preset. In other words, simply installing the new driver does not automatically switch every game to FSR 4.1. Whether it is actually available depends on the FSR version the game supports and the combination of GPU and driver you use.
Also, what has been extended to the Radeon RX 6000 series and various APUs this time is the upscaling feature only. Machine-learning-based frame generation, which creates intermediate frames, and ray regeneration, which reconstructs ray-tracing results, are not supported at the same time. AMD's official support table also clearly distinguishes these GPUs and APUs as "upscaling only." It is important not to take FSR 4.1 support to mean that all related features are now available.
This expansion lets users of older GPUs and handheld gaming PCs try the new machine-learning image reconstruction technology. However, the value of adopting it is not determined by peak frame rate alone. You need to compare image quality and speed in the same scene of the same game, and on handhelds match power settings as well, to see how much improvement you actually get. If FSR 4.1 can maintain satisfying image quality at lower internal resolutions than before, it may improve the balance between image quality and performance without buying a new GPU.