There is a price. With no dedicated AI cores, the near-six-year-old cards run the upscaler noticeably harder: frame rates were 13 to 34% slower than with FSR 3.1 across the tests. Similar gaps appeared when FSR 4.1 was tested on RX 7000 cards earlier this year, so the trade is a familiar one. The article's judgement is that the visual leap makes the sacrificed frames worthwhile, and that there is rarely any reason to keep using FSR 3.1 on these cards.

Against newer hardware the picture is more mixed. In Mafia: The Old Country and Horizon Zero Dawn Remastered, the RX 6950 XT's performance boosts relative to its native frame rate landed within 1 to 5% of the RX 7900 XTX, both cards running the same INT8 upscaling model. Cyberpunk 2077 stretched the gap to 10 to 31% in the newer card's favour, widening further as the upscaler stepped down from Quality through Balanced to Performance. The RX 9070 XT ran a native FP8 version of the model, with game settings adjusted to keep base frame rates comparable across all three cards.

The rollout is broader than one GPU family: when AMD announced the early arrival on 21 October, it said FSR 4.1 had been optimised for the entire RX 6000 range, from the RX 6950 XT down to the entry-level RX 6400, and was also coming to its APU integrated graphics and, with Microsoft, the Xbox Series X and S consoles. AMD has separately shipped a hyper-optimised version for RDNA 3.5 handhelds such as the ROG Xbox Ally X, though that release is due its own analysis. The Steam Deck, despite sharing RDNA 2 roots with the RX 6000 series, does not benefit, and there is no word on whether AMD can squeeze the model down for that system's chip.