NVIDIA DLSS 5 Review

DLSS 5 isn't just another incremental update — it's NVIDIA making a philosophical statement about where PC graphics are headed. Powered by what the company calls "Neural Rendering," DLSS 5 promises photo-realistic lighting and AI-generated image reconstruction that goes well beyond the upscaling tricks of previous generations. But as with every ambitious leap in tech, the reality is a little more complicated than the marketing slides.

What's Actually New Here

The headline feature is neural lighting — DLSS 5's ability to reconstruct physically plausible lighting from sparse ray-traced data. Digital Foundry's analysis of Resident Evil 9 showed striking before/after comparisons of Grace's character model, where the gap between base texture quality and DLSS 5 output was remarkable enough to spark a genuine community debate: how much of that leap is improved lighting versus AI hallucination? That tension sits at the heart of what makes DLSS 5 fascinating and slightly unsettling at the same time.

The context also matters: DLSS 5 builds on the transformer-based architecture introduced in DLSS 4, and the newer preset models — particularly Preset L — are showing up as genuinely impressive in early testing. One user who spent 50+ hours with Arc Raiders on an RTX 5080 at 4K/240Hz concluded definitively that Preset L is the best way to experience that game at 4K, citing cleaner output and better 1% lows when paired with Frame Generation.

Preset Wars: L vs. M vs. Everything Before

If you're jumping into DLSS 5 (or the transitional DLSS 4.5 presets that precede it), understanding the preset landscape is non-negotiable. The community has done serious homework here, and the findings are nuanced.

• Preset L is the current community favorite for clean output — especially at Performance mode in 4K. Users testing Oblivion Remaster noted it was the cleanest preset overall, with fewer artifacts than Preset M at equivalent settings.
• Preset M is sharper and more detailed in raster scenarios, but multiple reviewers confirm it becomes universally worse when ray tracing is enabled — extra noise, grass shimmer, and foliage boiling are common complaints in Witcher 3, AC Shadows, and Monster Hunter Wilds.
• The older Preset K still holds up better than M in RT-heavy games, which is a frustrating regression story that suggests DLSS 5's newer models are still finding their footing with real-time lighting.

One particularly insightful Reddit post explained why: the new model operates in linear color space rather than logarithmic, which makes it less effective at dampening the inherent flicker of ray-traced lighting. It's technically a "purer" approach, but practical results in RT games can look shimmery and unstable. The old model blurred that away — the new one exposes it.

The Denoiser Problem You Need to Know About

Here's the buyer tip that won't appear in any product listing: if you're running Unreal Engine 5 games with DLSS 5's newer presets, the Lumen denoiser can cause RT artifacts that make the output look worse than it should. The fix — disabling the Lumen denoiser via UE5 ini file edits — is a workaround that genuinely transforms image quality. One user reported that after disabling in-game sharpening through the UE5 ini, Preset M looked "amazing." This is a software/developer configuration issue more than a DLSS flaw, but it's real friction that buyers on RTX 50-series cards will encounter.

Who Actually Benefits from DLSS 5?

DLSS 5 is exclusive to RTX 50-series hardware. If you're on a 40-series card, you're capped at DLSS 4.x — still excellent, but the Neural Rendering features are off the table entirely. The performance story for RTX 5060 Ti (the most accessible Blackwell card) shows the card landing roughly where the 4070 did in raster performance, with DLSS 5 doing the heavy lifting to make it competitive at higher resolutions. That's the honest proposition: you're buying into AI-assisted performance scaling as much as raw GPU power.

For 4K enthusiasts with an RTX 5080 or 5090, the results are more compelling. Running Preset L at Performance mode — internally rendering at 1080p and reconstructing to 4K — users are reporting image quality that rivals or exceeds native TAA in many scenarios. The comparison of DLDSR 4K downsampled with Ultra Performance mode was described by one tester as looking "better overall than Native TAA 1440p," which is the kind of statement that would have seemed absurd two years ago.

Honest Caveats

The skeptics aren't wrong to pump the brakes. Some community members are calling DLSS 4.5/5 a near-sidegrade rather than a generational leap — "diminishing returns kicking in" is how one commenter put it. The fact that Preset M regresses in RT scenarios while Preset L lacks the same level of fine detail at distance objects (notable in Kingdom Come Deliverance 2) means there's no single "best" setting that works universally. Game-by-game, scene-by-scene tuning is still part of the reality.

The AI hallucination question is also legitimate. When DLSS 5 reconstructs lighting that wasn't present in the original render buffer — even if it looks better — there's a philosophical question about what you're actually seeing. For competitive gaming this might matter. For cinematic single-player experiences, most people will simply not care.

Frequently Asked Questions

Q: Does DLSS 5 work on RTX 40-series cards?

A: No. DLSS 5's Neural Rendering features are exclusive to RTX 50-series (Blackwell) hardware. RTX 40-series users can access DLSS 4 and the newer transformer-based presets, but not DLSS 5's photo-realistic lighting reconstruction.

Q: Which DLSS 5 preset should I use?

A: It depends on your use case. Preset L is currently the community favorite for clean output at 4K Performance mode, especially in non-RT games. In ray-tracing-heavy games, Preset M can introduce shimmer and foliage artifacts — the older Preset K may actually produce more stable results in those scenarios.

Q: Does DLSS 5 work with Frame Generation?

A: Yes, and the combination is powerful. Testing with NVIDIA Reflex showed it works best when Frame Generation is enabled — without FG, Reflex was observed to hurt 1% and 0.1% lows in some configurations.

Q: Is DLSS 5 better than FSR 4 or XeSS?

A: Based on current testing, DLSS 5 on supported hardware leads the pack for image quality, particularly in its Neural Rendering capabilities. However, FSR 4 runs on any GPU and has closed the gap significantly. The advantage of DLSS 5 is real but requires specific hardware to access.

Q: Are there any issues with DLSS 5 in Unreal Engine 5 games?

A: Yes — the Lumen denoiser in UE5 games can cause artifacts when combined with newer DLSS presets. Disabling the Lumen denoiser via ini file edits is a community-recommended workaround that meaningfully improves image quality in affected titles.

DLSS 5 is the most technically ambitious version of the technology yet — and the most conditional. If you have the hardware and the patience to dial in per-game settings, it delivers results that genuinely push the limits of what upscaled rendering can look like. If you're expecting a universal "always better" upgrade, you'll be doing more troubleshooting than expected.

— Tech Lead Editor, CPrice