DLSS vs FSR vs XeSS: How AI Upscaling Actually Works in Modern Games
If you've shopped for a graphics card or a gaming laptop recently, you've seen the acronyms: DLSS, FSR, XeSS. All three promise the same basic trick — render a game at a lower internal resolution, then use an algorithm to reconstruct a sharper, higher-resolution image, so you get frame rates closer to what the lower resolution would deliver with visual quality closer to native. But the three technologies are built very differently, run on different hardware, and produce noticeably different results in practice.
DLSS: Nvidia's hardware-accelerated approach
Nvidia's DLSS (Deep Learning Super Sampling) is the oldest of the three mainstream implementations and the one most tied to specific hardware. It runs on dedicated Tensor cores present only in Nvidia RTX graphics cards, and uses a neural network trained on extremely high-resolution reference images to reconstruct detail that a purely mathematical upscaler would miss — recovering fine textures, thin geometry like fences and power lines, and reducing shimmering in motion far more convincingly than older techniques. Because the model runs on dedicated silicon rather than borrowing from the same shader cores doing the rendering, DLSS's image reconstruction has historically carried less of a performance penalty for the quality it delivers, and successive versions have narrowed the gap with native rendering to the point where DLSS Quality mode is frequently indistinguishable from native resolution in motion, and sometimes even cleaner due to its built-in anti-aliasing effect.
The tradeoff is exclusivity: DLSS only runs on RTX-series Nvidia cards, and its highest-quality modes are further gated to newer RTX generations because the underlying neural model has grown more sophisticated with each release. If you don't own an RTX card, DLSS simply is not available to you in any form.
FSR: AMD's open, hardware-agnostic alternative
AMD's FidelityFX Super Resolution (FSR) took a deliberately different path: it's an open standard that runs on essentially any modern GPU, including Nvidia and Intel cards, older AMD hardware, and both current-generation consoles, because early versions relied on optimized spatial and temporal algorithms rather than requiring dedicated AI acceleration hardware. This is a major practical advantage for anyone without the latest, most expensive card — FSR support means an upscaling boost is available even on a mid-range or several-generations-old GPU. Newer FSR versions have moved toward machine-learning-based reconstruction similar in spirit to DLSS, narrowing the historical image-quality gap, though independent testing has generally still found DLSS slightly ahead in fine detail reconstruction and stability in fast motion, particularly in earlier FSR versions.
Because FSR is hardware-agnostic and open, it has also become the default upscaler on both major consoles for games that support it, which makes it relevant even if you never touch a gaming PC.
XeSS: Intel's middle path
Intel's XeSS is the newest of the three and takes a hybrid approach: it runs best with hardware acceleration on Intel Arc graphics cards, using matrix engines similar in concept to Nvidia's Tensor cores, but also ships a fallback mode that runs on general-purpose shader hardware for non-Intel GPUs, trading some quality and performance for broader compatibility. In practice, XeSS on an Arc card tends to land closer to DLSS in quality than FSR does, while XeSS running its fallback mode on non-Intel hardware performs closer to FSR. It's a genuinely clever compromise for Intel, which needed a competitive upscaler to make its graphics cards viable but doesn't yet have the installed base to make an Arc-exclusive feature attractive to game developers.
Why this matters more than raw GPU specs now
Upscaling technology has become central enough to how modern games are optimized that many titles are effectively designed with it turned on by default, particularly for ray-traced lighting, which remains extremely demanding to render at native resolution. This changes how you should think about buying a graphics card or gaming laptop: raw rendering performance at native resolution still matters, but which upscaling technologies a given GPU supports well, and how many of the games you actually play implement them, has become just as important a factor. A card with slightly weaker native performance but excellent DLSS support can outperform a nominally faster card stuck with a weaker upscaler in real, played frame rates. For related buying context, see our comparisons of PC gaming versus console costs and whether the Steam Deck OLED's screen upgrade is worth it, both of which touch on how modern hardware trades resolution for frame rate in practice.
Frame generation adds another layer
Layered on top of upscaling, all three vendors now also offer some form of frame generation, where the system inserts entirely AI-generated interpolated frames between rendered ones to multiply the displayed frame rate further. This is a genuinely different technique from upscaling — it doesn't reconstruct more detail, it manufactures additional frames — and it introduces its own tradeoffs, including added input latency that upscaling alone does not, since the generated frames need real frames on both sides before they can be created. Most implementations pair frame generation with a low-latency mode to offset this, but competitive multiplayer players in particular should test it carefully rather than assume more displayed frames always means a better experience.
It's also worth understanding that frame generation and upscaling are frequently marketed together under a single combined multiplier — a card advertised as delivering several times the base frame rate is usually stacking both technologies rather than showing raw rendering improvement, and that combined number can be genuinely misleading if you're trying to estimate actual responsiveness rather than just the number on the frame counter. For competitive, latency-sensitive games, it's worth benchmarking upscaling alone, without frame generation layered on top, to get a realistic sense of how a card performs when responsiveness matters as much as the visual frame rate.
Which one should guide your purchase?
If you're buying a new GPU or gaming laptop, don't just check whether a card "supports upscaling" — check which specific technology it runs natively and well, and cross-reference that against the games you actually play, since support and quality both vary title by title. Nvidia RTX cards get the most consistent, highest-quality experience today through DLSS, at a real price premium. AMD and console hardware get FSR's broad compatibility without requiring the newest silicon. Intel Arc sits in between. None of the three is a placebo — all genuinely improve frame rates with a real, if imperfect, image-quality tradeoff — but the "best" one for you depends entirely on which GPU you already have or are choosing between.
