In brief: AMD’s new FSR 4 upscaling solution improves tremendously upon FSR 3 but only supports the company’s Radeon RX 9070 and 9060 graphics card families. However, the latest version of Mesa for Linux provides a loophole that allowed a modder to benchmark FSR 4 on an older GPU, with intriguing results.
Redditor Virtual-Cobbler-9930 recently shared encouraging data after testing AMD’s FSR 4 on a Radeon RX 7900 XTX, which the new upscaling technology doesn’t officially support. While the FSR 4 hack incurs a significant performance penalty compared to FSR 3, it substantially improves image quality and still achieves higher frame rates than native resolution.
FSR 4, which switches from FSR 3’s spatial method to a machine learning method resembling Nvidia’s DLSS, requires FP8 computation, which AMD introduced with the RDNA 4 architecture in its RX 9070 and 9060 graphics cards. A recent update to Mesa for Linux enables FP8 emulation through FP16, which the earlier RDNA 3-based RX 7000 GPUs support.
FSR4 on RDNA3 (7900xtx) tests
byu/Virtual-Cobbler-9930 inradeon
From there, Virtual-Cobbler replaced the upscaling instructions in a few games using Optiscaler, a program that can hack FSR 4 into any title that supports DLSS 2, FSR 2, or XeSS. Currently, FSR 4’s biggest weakness is its limited official adoption compared to DLSS, but Optiscaler extends the functionality to over 200 additional games, making it highly useful for RDNA 4 GPU owners. While other methods exist, Optiscaler is likely simpler.
Also Read: AMD FSR 4 is Very Impressive: 1440p Upscaling Tested
Since the FP8 emulation loophole currently only works in Linux, Virtual-Cobbler tested it by running Windows games through the Proton compatibility layer on Arch Linux, the same technology that underpins SteamOS. While FSR 4’s performance cost on RDNA 3 in Windows remains unclear, like-for-like comparisons in Linux still show clear benefits.
In 4K with quality upscaling at maximum settings without ray tracing, the Cyberpunk 2077 benchmark reported 56 frames per second with FSR 4 compared to 85fps with FSR 3. However, Virtual-Cobbler believes that the image quality improvement justifies the roughly 20fps sacrifice. Furthermore, FSR 4 still provides noticeable performance savings, as native 4K drags the average framerate down to 42fps. Oblivion Remastered under similar conditions shows a 10fps difference between FSR 3 and 4, substantially improving motion clarity in exchange.
The downside is that the performance benefits of modding FSR 4 into RDNA 3 significantly diminish at lower resolutions. While the test suggests that scenarios exist where AMD could justify officially bringing its latest upscaling technology to older GPUs, those situations would be limited.
