What changed, and for whom
If you own a Series X or S, nothing changes on your console today. VideoCardz stresses that the announcement concerns developer access, not an immediate update for existing console games. Studios have to integrate the upscaler into each title before players see any benefit. VideoCardz also notes that AMD has not said which Xbox games will support FSR 4.1.2 first. A separate report, from Archyde, also says Microsoft has published no list of compatible games or dates for first updates.
Previously, developers using the analytical FidelityFX Super Resolution on Xbox now have an upgraded option for higher perceived resolution and image quality. Analytical means the older, non-ML approach. The new version is neural.
Two models: Full INT8 and Lite INT8
There are two models, and both run on Series X and Series S. They are Full INT8 and Lite INT8. AMD says Lite is less than half the size of Full and has a lower upscaling cost. Developers pick the balance of GPU cost and image quality that suits each game.
Memory is not part of that trade-off. AMD says upscaler memory is the same for both models because it depends on render and display resolution, not on the model. The GDK sample can switch between Full and Lite at runtime, so developers can compare both in the same scene.
Microsoft's GPU-cost figures
Microsoft's Xbox developer post lists GPU time from its GDK sample. It says the Xbox team optimized FSR Upscaling 4 for the consoles, making ML-based upscaling practical on their RDNA 2-based GPUs.
| Console | Model | Output resolution | GPU time |
|---|---|---|---|
| Series X | Lite | 1440p | 1.29 ms |
| Series X | Lite | 4K | 2.75 ms |
| Series X | Full | 1440p | 1.95 ms |
| Series X | Full | 4K | 3.86 ms |
| Series S | Lite | 1080p | 1.9 ms |
| Series S | Lite | 1440p | 3.4 ms |
| Series S | Full | 1080p | 2.9 ms |
| Series S | Full | 1440p | 5.2 ms |
These are sample measurements of the upscaling pass, not game benchmarks. They are not a promised frame-rate gain. A 60 fps game has a frame budget of about 16.7 ms. Full at 4K on Series X takes 3.86 ms of that, while Lite takes 2.75 ms. That is a general calculation on my part, not something Microsoft states. Whether the quality gain justifies the cost will depend on each title.
Series S shows the squeeze most clearly. Full at 1440p costs 5.2 ms, and Lite costs 3.4 ms. Lite looks like the natural fit for the smaller console, but that is my inference. Studios will decide for themselves.
Microsoft's GDK notes make a further claim. They say the Xbox-specific optimizations give higher upscaling throughput per teraflop than RDNA 2 PC GPUs running the same Full INT8 model. That is Microsoft's own comparison, and I have seen no independent testing of it.
Integration: same SDK as Windows
For developers, the work should look familiar. The Xbox version uses the same AMD FSR SDK API as Windows. Registered Xbox developers get AMD-signed binaries for Series X|S in the GDK. It takes the same inputs developers already provide for FSR on PC: jittered color, depth and motion vectors. The reactive mask and the transparency and composition mask are optional.
The quality modes also match PC. They are Native AA, Quality, Balanced, Performance and Ultra Performance. AMD recommends using the standard names and scaling ratios in game settings, so players get consistent controls across titles.
Microsoft's GDK sample builds for both Xbox Series X|S and Windows desktop. One report says it implements FSR 4.1.2 and requires GDK October 2025 or later. The same report says the sample falls back to FSR4 Full if no FSR 3 provider is available.
Handhelds are a separate path
The ROG Xbox Ally family is covered too, but through a different route. AMD says these are Windows devices, so developers use the standard FSR SDK. SDK 2.4 supports the Ryzen AI Z2 Extreme and Ryzen Z2 A processors, using an efficient INT8 model. An ASUS preview driver for the Ally X and X20 upgrades DirectX 12 games already using FSR 3.1 or newer. Support for the standard Ally follows in a later update. AMD recommends integrating the latest SDK natively. It says that is also required for FSR Upscaling 4.1.2 on Vulkan.
Why INT8
TechPowerUp adds some hardware background. RDNA 4 handles 8-bit floating point natively, so FSR 4.1 runs on FP8 there. The RDNA 2 GPU IP in the Xbox consoles has no FP8 support and uses INT8 data types. That explains why the Xbox models are INT8, and why Lite matters on older hardware.
What to watch
- First shipping games. None are named yet.
- Real-world cost. Look for developer or press comparisons of Full and Lite against FSR 3 in actual games.
- Player-facing options. Studios will decide whether to expose the new modes in their settings menus.
The practical takeaway is that Xbox now has a Windows-compatible ML upscaling path with a cheaper option for the weaker console. Players will see it only when studios ship it.