PC GamingGraphics TechPerformance Tuning

DLSS vs FSR vs XeSS: The Differences That Actually Matter

By Wendell Al-Hassan · August 23, 2026 · 12 min read

Upscalers have become the default way to run modern games with high settings and ray tracing. You’ve probably searched DLSS vs FSR and ended up with absolutist takes that don’t help in the moment you actually need to pick a setting. The real answer depends on which version you have, your GPU, your monitor’s refresh, the game engine, and what kind of artifacts you notice most.

This piece is a practical comparison of NVIDIA’s DLSS (Super Resolution, Frame Generation, Ray Reconstruction), AMD’s FSR (2/3 and Native AA), and Intel’s XeSS. It focuses on what changes your experience on screen: motion stability, detail retention, latency, and the trade‑offs that appear at 1080p, 1440p, and 4K.

How modern upscalers work (in one usable paragraph)

Most current solutions are temporal upscalers. Instead of guessing a high‑res image from a single low‑res frame, they use motion vectors, depth buffers, and a history of jittered frames to reconstruct detail while acting as an anti‑aliasing pass. In practice, that means the quality you see depends as much on the game’s motion vectors and how the engine handles disocclusion (newly revealed pixels after movement) as on the brand of the upscaler. A clean HUD and sharp static screenshots tell you little; watch foliage, specular highlights, thin geometry, and fast camera pans to judge properly.

Hardware support and where each runs

• DLSS Super Resolution and Frame Generation require RTX GPUs with Tensor cores. Super Resolution works on RTX 20/30/40 series; Frame Generation is limited to RTX 40 series in games that implement it. DLSS Ray Reconstruction also runs on RTX hardware in supported titles.
• FSR 2/3 runs on a wide range of GPUs from all vendors because it doesn’t require dedicated AI hardware. Its Frame Generation can work on many modern GPUs, with some games allowing vendor‑agnostic FG; AMD also offers AFMF (driver‑level frame generation) for DX11/12 in windowed/borderless modes.
• XeSS supports two paths: an AI‑accelerated XMX mode on Intel Arc GPUs and a DP4a path for NVIDIA and AMD cards without XMX. Quality is best on Arc hardware but widely usable elsewhere.

That matrix matters because it sets your ceiling. If you don’t have an RTX 40 card, you can’t use DLSS Frame Generation in games. If you’re on an older NVIDIA or AMD GPU, you can still run FSR 2/3 and XeSS in many titles, and sometimes they’ll look close to vendor solutions if the game’s implementation is good.

DLSS vs FSR: where each wins in image quality

• Temporal stability in motion: DLSS Super Resolution (2.x) tends to hold fine detail and specular highlights better, especially at lower internal resolutions. Edges on wires, distant foliage, and moving transparencies (like fences) usually shimmer less. FSR 2/3 has improved, but it can still exhibit more flicker or crawling in sub‑pixel detail, and more visible break‑up during fast pans.
• Ghosting and trails: All temporal solutions can ghost when motion vectors are wrong or materials are tricky (particle effects, water, moving alpha). DLSS frequently leaves shorter, less opaque trails; FSR may show longer ones in some games, particularly around bright HUD elements or thin effects. Implementation matters: some Unreal Engine titles with well‑fed motion vectors narrow the gap.
• Disocclusion handling: DLSS often rebuilds newly exposed pixels with fewer sparkly artifacts; FSR can show grain or “sticky” pixel noise after fast motion. XeSS sits between them—better than early FSR in many cases, not as consistent as DLSS at low render scales.
• Built‑in sharpening: DLSS’s slider is conservative and less prone to haloing. FSR’s sharpening (RCAS) is stronger by default; it can make edges pop but can also accentuate noise. Dial it down if you see halos around UI or bright geometry.

The shorter version: if you have the choice and are sensitive to shimmer and ghosting, DLSS Super Resolution is usually the safer pick. If DLSS isn’t available, FSR 2/3 and XeSS can be very usable, with XeSS sometimes delivering a slightly cleaner result than FSR at the same render scale, depending on the game and your GPU path.

Picking the right quality mode by resolution

A common mistake is using the same upscaler preset across resolutions. Internal resolution matters more than the label.

• 4K displays:
  • DLSS Quality or Balanced often look near‑native while boosting performance significantly. DLSS Performance can still be acceptable for cinematic titles but may show loss of microdetail.
  • FSR 2/3 Quality is the best default. Balanced is workable for action titles; Performance starts to show more flicker in distant detail.
  • XeSS Quality is a good starting point; move to Balanced only if you need the frames.
• 1440p displays:
  • DLSS Quality is the sweet spot; Balanced when GPU headroom is tight. Performance mode becomes more obviously soft or unstable.
  • FSR 2/3 Quality is recommended. Balanced is serviceable but watch for shimmer; Performance is usually too compromised.
  • XeSS Quality typically; consider Balanced only if ghosting remains controlled.
• 1080p displays:
  • Upscaling can work, but artifacts are most visible. Prefer DLSS Quality (or DLAA if you don’t need the performance) over any lower mode.
  • FSR and XeSS at 1080p Quality can be okay in slower games; many players prefer native with good TAA or vendor “Native AA” modes here.

When in doubt, switch to a test scene with sharp diagonals, foliage, and specular detail, then toggle modes while panning the camera slowly. It’s the most honest way to see the trade‑offs.

Frame Generation: speed, smoothness, and what “fake frames” change

• How it works: FG synthesizes intermediate frames using optical flow and motion vectors, roughly doubling displayed frame rate without doubling input samples from the game. DLSS 3’s FG leverages dedicated hardware on RTX 40; FSR 3’s FG works more broadly and some titles expose it independently of the upscaler.
• Latency: Displayed FPS goes up; input latency doesn’t drop in the same proportion. NVIDIA Reflex integration offsets some latency by limiting render queueing. AMD recommends pairing FG with Anti‑Lag or the game’s low‑latency mode. You still shouldn’t expect 60‑to‑120 FG to feel like true 120 in fast shooters.
• Visual stability: FG can exhibit artifacts around moving UI, alpha effects, or fast motion where optical flow struggles. DLSS FG tends to preserve object boundaries slightly better in many titles; FSR FG quality varies more by game.
• Driver‑level FG: AMD AFMF injects FG at the driver level for many DX11/12 titles in borderless/windowed. It’s convenient but more susceptible to HUD artifacts and cadence issues than per‑game implementations. NVIDIA does not provide a general driver‑level FG equivalent in current public drivers; DLSS FG is integrated per‑title.

Use FG for cinematic, single‑player games, racing, and flight sims where extra smoothness helps with camera motion and the latency trade is acceptable. Avoid FG in twitch shooters or competitive play unless you test and genuinely can’t feel a penalty.

Latency tools and how to balance smoothness

• Reflex (NVIDIA): Reduces render queue latency and pairs well with DLSS FG. Enable it when available; combine with a frame cap slightly below your monitor’s max to prevent VRR from drifting into poor frametimes.
• Anti‑Lag / Anti‑Lag+: AMD’s latency reduction. Works with FSR and, in many cases, with FG. Recent driver changes have adjusted support in certain anti‑cheat environments; use the game’s built‑in low‑latency options when in doubt.
• Frame capping and VRR: With FG active, cap to a value your system can maintain without oscillation (for instance, 5–10% below VRR ceiling). It reduces microstutter and keeps latency stable.

If you’re CPU‑bound, FG can still improve perceived smoothness by filling in. It will not fix input sampling; the game is still updating at the CPU‑limited rate. That matters in busy city hubs or strategy games with lots of simulation.

DLAA, FSR Native AA, and when native is better

Sometimes you don’t need to upscale at all.

• DLAA: NVIDIA’s DLSS network used purely as an anti‑aliasing filter at native resolution. It’s excellent for shimmer‑prone content and pairs well with high‑end GPUs at 1440p/4K when performance is already sufficient.
• FSR Native AA: Available in some games. Applies FSR’s anti‑aliasing and reconstruction at native res without scaling. Results vary; it can be cleaner than the game’s TAA but rarely matches DLAA’s temporal stability.
• XeSS at native: A few titles allow XeSS at scale 1.0. Quality can land between good TAA and DLAA depending on implementation.

If you’re on a 1080p or 1440p high‑refresh display and already hitting your target FPS, these “native AA” modes can clean edges with fewer reconstruction side effects.

Engine quirks and implementation quality

• Unreal Engine: Historically, Unreal’s built‑in TAA and TSR (Temporal Super Resolution) vary by version. Good motion vectors and ghost rejection can make FSR/XeSS close the gap to DLSS. Bad vectors create halos and trails across all vendors.
• Motion blur and film grain: These can mask shimmer but also hide detail and make sharpening look worse. If your upscaled image looks noisy, try disabling film grain first, then tweak sharpening.
• UI/HUD: Upscalers that touch the entire frame can produce soft HUDs. Many games draw UI at native res; use that option when available. If the HUD shimmers with FG on, look for a “UI render at native” or “UI FG mask” toggle.
• TAA sharpening vs upscaler sharpening: Avoid stacking both. If the game has a separate TAA sharpener and an upscaler sharpen, pick one to prevent halos.

The bottom line on implementations: when two titles use the same upscaler, they still won’t look identical. Blame (or credit) the engine’s data quality as much as the upscaler brand.

Practical presets by game type

• Competitive shooters (1080p–1440p, high refresh):
  • Prefer native or DLAA if frames are ample. If you must upscale, DLSS Quality or FSR Quality, minimal sharpening, Reflex/Anti‑Lag on, FG off. Cap frames to reduce latency spikes.
• Open‑world action RPGs (1440p–4K):
  • DLSS Quality/Balanced or FSR Quality; consider FG if traversal feels uneven and input latency remains tolerable. Tweak sharpening based on foliage shimmer.
• Racing/flight sims:
  • Upscaling + FG often feels great. DLSS Balanced with FG is a common happy medium; FSR Quality/FG works well too. Watch instrument panels for ghosting; adjust sharpening to keep gauges legible.
• Strategy and city builders:
  • These are often CPU‑limited. Upscaling to free GPU headroom helps little; FG can still improve perceived scrolling smoothness but won’t speed simulation ticks. Prefer quality modes to preserve small text clarity.

1080p vs 1440p vs 4K: artifact visibility and expectations

• 1080p: Internal render resolutions in Performance/Ultra Performance modes are simply too low. Temporal solutions struggle with sub‑pixel detail, leading to crawl. Stay on Quality (or native) and keep sharpening gentle.
• 1440p: The balance point. Quality modes are visually strong; Balanced is often okay. If you’re sensitive to shimmer, prioritize DLSS Quality or DLAA if you can spare the frames.
• 4K: Upscaling shines here because internal resolutions remain decently high even in Balanced modes. Quality differences compress; many players happily run DLSS Balanced or FSR Quality for heavy ray tracing.

When ray tracing enters the chat

Ray tracing pushes GPUs hard and changes the calculus.

• DLSS Super Resolution is often the difference between playable and not when multiple RT effects stack. Use DLSS Balanced at 4K or Quality at 1440p to keep RT on without tanking frames.
• FSR 2/3 helps similarly, though you might need to stick to Quality at 1440p to avoid shimmer. FG on top can make RT‑heavy scenes feel fluid, but watch for artifact amplification in glossy areas where optical flow is tricky.
• DLSS Ray Reconstruction (in supported titles) replaces hand‑tuned denoisers with a learned model. It can reduce noisy reflections and bounce lighting flicker. When available, pairing RR with DLSS SR tends to improve both clarity and stability in RT scenes.
• XeSS with RT: Viable but more sensitive to motion vector quality in reflections. Judge per game; many look fine at Quality with moderate RT settings.

If your card can’t sustain RT even with upscaling, consider mixing: turn down the most expensive RT pass (often global illumination or multi‑bounce reflections) before lowering the upscaler mode.

Troubleshooting common artifacts (fast fixes that work)

• Shimmering foliage and power lines:
  • Raise the upscaler mode (Performance to Balanced, Balanced to Quality).
  • Reduce sharpening; disable film grain.
  • If available, switch to DLAA at native for dense foliage scenes.
• Ghost trails on moving objects:
  • Lower sharpening; it can exaggerate trails.
  • Check for motion blur settings that help blend minor ghosts without smearing.
  • Some titles offer “anti‑ghost” toggles; enable them even if they slightly soften the image.
• Soft text and UI:
  • Ensure UI is set to render at native res.
  • Use the upscaler’s sharpening instead of a separate TAA sharpen.
• Sparkly disocclusion when rotating the camera:
  • Bump upscaler quality one notch.
  • Cap frames to a stable value to reduce cadence irregularity that emphasizes sparkle.

Driver‑level spatial upscalers and when to use them

• NVIDIA Image Scaling (NIS): A spatial scaler and sharpener you can force from the control panel. It lacks temporal data, so it won’t fix shimmer like DLSS, but it’s universal and low overhead. Good for older titles without built‑in upscalers.
• Radeon Super Resolution (RSR): Similar to NIS but integrated with AMD drivers, upscaling the entire frame. Works for many games, but HUD/text can appear softer since it’s a post‑process.
• Where they fit: Use when a game lacks DLSS/FSR/XeSS and you simply need more frames. Prefer in‑game temporal upscalers whenever available—they reconstruct detail with history and motion awareness in a way spatial scalers can’t.

A clean way to decide settings in any new game

• Start native, no sharpening. Note baseline quality and FPS in a scene with foliage, fine geometry, and motion.
• Enable DLSS vs FSR vs XeSS if available, set to Quality. Toggle while panning to judge shimmer and ghosting.
• If 4K and GPU‑limited, try Balanced next; at 1440p, only move to Balanced if the Quality mode’s FPS is still below target.
• Consider Frame Generation only after you’ve met a minimum true FPS (often around 50–60) and input latency feels acceptable with Reflex/Anti‑Lag on.
• Cap frames just under your display’s ceiling; verify frametime stability in motion, not just the average FPS counter.

Two minutes of disciplined testing beats hours of guessing.

Advanced note: engine‑native solutions and the role of Ray Reconstruction

Engine‑native temporal solutions like Unreal Engine’s TSR and modern in‑house TAAU continue to improve. In some titles, TSR on its “High” setting lands close to DLSS Quality; in others, it exhibits more shimmer than vendor upscalers. The advantage of vendor tech grows when ray tracing is heavy, because denoising becomes the bottleneck and the interaction between the upscaler and the ray‑traced signal quality is critical.

This is where DLSS Ray Reconstruction changes the stack in supported games. By replacing multiple hand‑tuned denoisers with a learned model that understands temporal context, RR can clean up reflections and global illumination more consistently, giving the upscaler a better input to work with. The benefit isn’t universal—if RT is light or the scene is mostly diffuse, the win may be small—but when glossy, noisy content dominates, RR plus DLSS SR often produces a steadier image than traditional denoisers paired with any upscaler. As developers ship more titles with decoupled upscalers and FG (as seen with newer FSR 3 integrations) and richer motion data, expect the “vendor vs vendor” debate to matter less than whether the game feeds the algorithm clean information in the first place.