NVIDIA Pulls Ahead Again: DLSS 5 Does What FSR 4 Can't (Yet)

Over the past two years, AMD has practically closed the gap with NVIDIA in image upscaling. According to TechPowerUp, FSR 4 has reached a level of image quality comparable to DLSS, but the arrival of DLSS 5 has widened the gap once again — and now it's less about upscaling quality and more about the fundamental approach to rendering itself.

NVIDIA first unveiled DLSS 5 back in March at GTC, calling it the biggest change in real-time graphics since the introduction of ray tracing. The community reaction was cautious: people feared the AI would start redrawing frames on its own, overriding the artists' original vision. Unlike previous versions, DLSS 5 isn't about reconstructing an already rendered image — it adds new details on top of it: photorealistic lighting, subsurface scattering on skin, contact shadows, fabric sheen. The model can layer light and materials over the frame, but it leaves the geometry untouched — silhouettes, normals, and shapes remain exactly as the engine rendered them. Developers can tweak the intensity of these effects for different parts of the image or disable DLSS 5 entirely for specific objects.

FSR 4 and DLSS 5 now have fundamentally different goals. FSR 4 is classic upscaling: the algorithm takes a low-res frame and tries to reconstruct it at a higher resolution. DLSS 5, meanwhile, is neural rendering — it generates entirely new visual information on top of an already finished frame. AMD doesn't have an equivalent tech yet, and neither does Intel. On the integration side, basic DLSS 5 implementation in games doesn't require specialized render passes — the tech just needs the finished frame and motion vectors. For the best results, developers will need to fine-tune how it handles characters, environments, and objects separately, but even without that tuning, DLSS 5 will still work out of the box.

The ease of implementation means studios could start cutting corners on the raw quality of character assets. Motion capture and high-poly models are expensive, and some of that visual detail can now be added post-render via AI. However, the tech is computationally heavy — according to NVIDIA, enabling DLSS 5 on GeForce RTX 50 cards drops performance by an average of 50–60%, depending on resolution. The company notes that the performance hit doesn't vary much from game to game — the model isn't tuned per title.

DLSS 5 runs exclusively on GeForce RTX 50 graphics cards with fifth-gen Tensor cores. When asked whether RTX 40 and older cards might at least get the option to run the tech (even with poor performance), NVIDIA gave a vague response: "DLSS 5 is the largest and most complex model we've built, and our priority was optimizing it for RTX 50." Still, enthusiasts have already managed to run a leaked test build of DLSS 5 on RTX 40 hardware — Ada cards are technically capable of handling these models. The real question is whether NVIDIA will officially allow it.

AMD has its own momentum — FSR 4, launched in early 2025 as an RDNA 4 exclusive, now supports over 300 games by mid-2026. In July 2026, the tech started rolling out to RDNA 3 (Radeon RX 7000), and RDNA 2 (RX 6000) owners are promised support in early 2027. In terms of image quality, FSR 4 in Quality mode is now comparable to DLSS in most scenarios, though in blind tests, DLSS 4.5 still pulls about 48% of the votes compared to FSR 4's 15%. FSR 4.1, which RDNA 3 users will receive, has noticeably improved sharpness and noise reduction.

This feels like a race where AMD nearly caught up on the straightaway — only for NVIDIA to veer onto a completely different track. FSR 4 solved the upscaling problem admirably, but NVIDIA is already playing a whole new game: neural rendering, where its competitors don't even have a starting position yet.

Do you think AMD can catch up to NVIDIA in neural rendering, or should they focus on evolving FSR in the direction where they're already seeing results? And would you be willing to sacrifice 50% of your performance for a more detailed AI-enhanced image? Drop your take in the comments.