Thursday, May 14th 2026
AMD FSR Upscaling 4.1 Officially Arrives to Radeon RX 7000/6000 GPUs
AMD has responded to what gamers have been most vocal about: bringing its FSR 4.1 upscaling technology to older RDNA 3 and RDNA 2 GPUs. Today, we learn that this dream is becoming a reality, as AMD is officially introducing its FSR Upscaling 4.1 technology to millions of Radeon RX 7000 Series graphics cards worldwide. Arriving this July, millions of RDNA 3 gamers will enjoy official support for FSR Upscaling 4.1 in over 300 supported games. This means that, just like the latest RDNA 4-based Radeon RX 9000 Series, RDNA 3 gamers will also benefit from the upscaling algorithms based on various machine learning techniques without needing to upgrade to the latest Radeon RX 9000 GPUs if their current GPU performs adequately.
Interestingly, the FSR Upscaling 4.1 will use the same underlying model on RDNA 3 as it does on RDNA 4. However, differences in hardware create some background variations, as RDNA 4 has FP8 processing units and INT8, whereas RDNA 3 only has INT8. AMD explains that the company has developed FSR Upscaling 4.1 with the FP8 format in mind, while the latest ML algorithm has been optimized and ported for the INT8 data format to work with RDNA 3. This difference in hardware support delayed AMD's port of FSR technologies to RDNA GPU IP below version four, but it is now finally making an appearance.
Finally, FSR Upscaling 4.1 is also coming to older RDNA 2 GPUs, but that port will have to wait until early 2027. This indicates that AMD is doing additional optimization for RDNA 2 cards, as it is challenging to port this upscaling algorithm to hardware that isn't optimal for the latest techniques. However, it is promising that it is possible, and RDNA 2 gamers will just have to wait a bit longer.
Source:
Jack Huynh on X
Interestingly, the FSR Upscaling 4.1 will use the same underlying model on RDNA 3 as it does on RDNA 4. However, differences in hardware create some background variations, as RDNA 4 has FP8 processing units and INT8, whereas RDNA 3 only has INT8. AMD explains that the company has developed FSR Upscaling 4.1 with the FP8 format in mind, while the latest ML algorithm has been optimized and ported for the INT8 data format to work with RDNA 3. This difference in hardware support delayed AMD's port of FSR technologies to RDNA GPU IP below version four, but it is now finally making an appearance.
252 Comments on AMD FSR Upscaling 4.1 Officially Arrives to Radeon RX 7000/6000 GPUs
Raw TOPS matter, but so do data formats, driver work, memory bandwidth, VRAM capacity and actual runtime overhead. This matters even more for weaker GPUs and iGPUs. This is complicated.
The technical limitations are real, but AMD’s poor communication still made the whole thing look worse than it needed to.
FSR 4.1 may technically run across RDNA 2/3, but practical usefulness will probably vary a lot. High-end RDNA 2 and most RDNA 3 cards seem like much more realistic targets for a good experience, while low-end 4 GB cards and iGPUs may be much more limited by overhead, bandwidth or VRAM.
I am happy that I've got RDNA4 with lots of TOPS and 16 GB, so updates keep coming and will probably be fully supported. I would say that RDNA4 is roughly on a similar feature-set level as Nvidia Lovelace.
Sure it may not make sense for every gpu and every situation but it's still just stupid segmentation not to release it. Even if it can be optimized to perform better.
Same could have been said about ray tracing on older amd cards, rdna2 barely made it better from a hardware standpoint for it. RTG's leadership really should get a big shakeup.
What scares me is the focus shifting from brute force raster gains to smarter rendering tricks and AI. If each new gen delivers solid native perf and holsters upscaling tech like an extra tool in the box, I'm all for it. What won't wash well is eventual next gen 10% uplift cycles (raw perf) paired with upscaling + x8 FG to compensate, unless these SMART techniques pull it off by mirroring native rendering.
RX 6400
RDNA2, 768 Cores @2300 MHz
4GB VRAM at 128 GB/s
RX 780m
RDNA3, 768 Cores @2700 MHz
8GB shared VRAM at 89 GB/s
In practice the 6400 is about 20% faster but with only 4GB this needs to be carefully managed in some games. Also the 780m can be configured to 4 or 16GB if needed but 8GB is ideal. It'll be interesting to see if and how INT8 works differently on these GPUs should both get the option.
Just because something doesn't affect you in a meaningful way (for now), you don't have to be negative about it.No, it's not like that at all.
The announcement doesn’t say a thing about integrated graphics. It specifically calls out 6000 and 7000 series discrete GPUs.
The PS5 Pro has 300% of the TOPS performance of a 7900XT.