DLSS vs FSR vs XeSS: What Gamers Need to Know About Upscaling
Upscaling has become a standard part of the graphics menu, but DLSS, FSR and XeSS are not interchangeable. This guide explains how each approach works, which GPUs can run it, and how to pick the right mode and quality preset for your hardware.
admin
Upscaling has quietly become one of the most consequential settings in a modern PC game. In the past, players who wanted more frames had two levers to pull: lower the resolution or lower the detail settings. Both made the image visibly worse. Upscaling offers a third path by rendering the frame at a lower internal resolution and then reconstructing it back up to your monitor's native resolution, which can produce a large jump in frame rate without the mushy look of a straight resolution drop.
All three major technologies in the market today aim at that same goal, but they get there in different ways and, crucially, run on different hardware.
How the three approaches differ
DLSS is Nvidia's system and depends on dedicated tensor cores inside GeForce RTX graphics cards. The technique is temporal: the renderer feeds in motion vectors and previous frames, and a trained neural network fills in the missing detail. Because it leans on hardware that only RTX cards have, DLSS is not available on Radeon or Intel GPUs, but on supported RTX hardware it has generally been the benchmark for image stability in motion.
FSR is AMD's answer, and its defining feature has historically been reach. Early versions used a spatial, non-AI upscaling method that could run on a very wide range of GPUs, including Nvidia and Intel cards, which is a large part of why FSR appeared in so many games ¹. Newer iterations have moved toward machine-learning reconstruction, but the compatibility story remains FSR's biggest selling point and the reason it is often the only upscaler available to someone on older or budget hardware.
XeSS is Intel's implementation and is the most hardware-agnostic of the ML-based options. On Intel Arc cards it uses dedicated XMX matrix hardware; on other GPUs it can fall back to a DP4a path that runs on more general-purpose shader hardware. That fallback usually costs some performance and image quality compared with the Arc version, but it means XeSS is not locked to a single vendor. Like DLSS, it reconstructs detail with a trained model rather than simple spatial interpolation ¹.
What this means in practice
The first rule of choosing an upscaler is not to choose at all: use the one your GPU is built for. On an RTX card, DLSS is the obvious first stop because it makes use of silicon you already paid for. On Radeon, FSR is the native path. On Arc, XeSS is the strongest option and Intel has put real effort into it. Only when your preferred technology is missing from a game should you start comparing alternatives on image quality alone.
The second rule is that naming has become confusing. Vendor version numbers, quality presets and frame-generation features are now layered on top of each other, and the same option label can behave differently depending on which GPU generation you own. Some newer DLSS transformer models, for example, are noticeably heavier on RTX 20- and 30-series cards because those GPUs lack hardware support for the lower-precision formats the newer models use, which can make an older model the smarter pick on older hardware ². The practical takeaway is to treat the version number in a review as a starting point rather than a guarantee about your own card.
Understanding the presets
Every upscaler ships with a handful of quality tiers, and these matter more than most players expect. Quality mode renders at a relatively high internal resolution and typically looks close to native at 1440p and 4K. Balanced sits in the middle. Performance and Ultra Performance render much lower and lean heavily on reconstruction, which is where artifacts, shimmering on fine detail, and unstable thin objects such as fences and foliage tend to appear.
Two factors decide how far down you can go. The first is your output resolution: a preset that looks acceptable at 4K may look rough at 1080p, because there are simply fewer output pixels to hide reconstruction errors. The second is your frame rate. Temporal upscalers have less information to work with when frames are short, and artifacts are far easier to spot while moving than in a static screenshot. Judging quality from a still image is one of the most common mistakes people make.
What to actually test
Load a demanding scene, enable the upscaler, and then move the camera. Look at thin geometry, distant text, particle effects and the edges of the HUD. Watch for ghosting trails behind fast objects, flickering on specular highlights, and smearing around characters. If those problems are distracting, step up one quality tier before you consider turning the feature off entirely.
Also be clear about what you are enabling. Super resolution upscaling, frame generation and latency-reduction technologies are separate features that often appear in the same menu. Frame generation inserts interpolated frames and improves smoothness rather than responsiveness, and it works best when your base frame rate is already reasonable. Stacking every toggle at once can produce a big number on the counter and a worse-feeling game.
The bottom line
DLSS, FSR and XeSS have converged enough that the honest answer is no longer "one wins everywhere." DLSS remains the strongest choice on RTX hardware, XeSS is impressively capable for a cross-vendor option, and FSR's wide compatibility keeps it relevant for the many players who are not on Nvidia. Start with the technology your GPU accelerates, pick the highest quality preset you can tolerate in motion, and treat frame generation as a separate decision from upscaling itself.
