Me: You can now run DLSS 5 on AMD GPUs. My RX 9070 XT: Please stop, it hurts us
A PC Gamer piece looks at community workarounds that let Nvidia's DLSS technology run on AMD graphics cards, using tools that swap in Nvidia's upscaling and frame-generation libraries even on hardware they weren't designed for. The article, written in a light-hearted, first-person tone, argues that while such swaps are technically possible, they come at a real cost: AMD cards lack the dedicated tensor hardware Nvidia's DLSS relies on, so forcing it to run produces heavy performance penalties rather than the smooth gains Nvidia GPU owners see.
The piece centres on the author's own AMD Radeon RX 9070 XT, used as a case study for why running DLSS 5 outside its intended Nvidia ecosystem is more a novelty than a practical upgrade. It frames the exercise as an illustration of the gap between "can be done" and "should be done," noting that without matching hardware acceleration, the workaround shifts extra strain onto the AMD GPU rather than delivering the efficiency DLSS normally provides on Nvidia cards.
In short
- DLSS 5 can technically be forced to run on AMD GPUs
- AMD cards lack hardware DLSS needs, hurting performance
- Author uses their RX 9070 XT as a cautionary example
Explained simply
DLSS is Nvidia’s technology for upscaling game images and generating extra frames, helping games look smoother or run faster on supported Nvidia graphics cards. It relies on dedicated hardware in those cards, while AMD graphics cards, including the Radeon RX 9070 XT, use different hardware and software.
Community tools can sometimes swap in Nvidia’s DLSS libraries on AMD cards, but the software cannot give those cards Nvidia’s dedicated hardware. Running it this way can place much more work on the AMD GPU, so the result may be a technical curiosity rather than a useful performance boost.
The issue matters because graphics features depend on both software and the hardware they were designed to use. A workaround that launches on another brand’s card does not necessarily deliver the same speed or efficiency.