DLSS 5: NVIDIA promises a graphic revolution, players cry scandal

March 16, 2026, GTC keynote. Jensen Huang takes the stage in his usual leather jacket and drops a bomb: DLSS 5 . Not just an update. Not a DLSS 4.5 with one more patch. No, a complete change of philosophy.

So far, DLSS was essentially doing two things: upscaling (rendering in low resolution and reconstructing in high resolution via AI) and frame generation (interpolating images calculated by AI between real ones). DLSS 4 with Multi Frame Generation already pushed this quite far on RTX 50. But DLSS 5? It's something else.

This time, AI is no longer content to just reconstruct or invent frames. modifies pixels directly to inject photorealistic lighting, more credible materials, micro-reflections, boosted ambient occlusion - in short, everything that makes the difference between real-time rendering and a Hollywood movie scene. NVIDIA calls it the neural rendering .

Jensen didn't hesitate to use strong language: "This is the GPT moment of video games." Just that.

The principle is quite simple to understand, even if the execution is a technical monster. The game engine does its classic rendering - polygons, textures, shaders, ray tracing if activated. So far, nothing new. But then, the DLSS 5 neural model analyzes the color buffers and motion vectors to identify each element of the scene: a character's skin, hair, a metallic surface, fabric, water.

And there, it reconstructs these elements with a level of detail that the engine alone cannot achieve in real time. We're talking about rim lighting on the hair, finer contact shadows, reflections on surfaces that normally approximate them just with cubemaps. All powered by NVIDIA's supercomputers.

The catch? For now, the demos were running on two RTX 5090 in parallel — one for gaming, one for AI model. Clearly not representative of what we will have at home this fall. NVIDIA promises optimizations, but we wait to see.

And that's where it goes off the rails.

The first demos triggered an almost unanimous reaction on the networks: "It's AI slop." The images shown — notably on Resident Evil Requiem and Hogwarts Legacy — have shocked quite a few people. Characters with skin that is too smooth, almost waxy. Lighting that crushes the original atmosphere of the game. And above all, an impression that all games end up looking the same: photorealistic, smooth, uniform.

The term that comes back the most? The "yassification" video games. You take a character with a specific artistic direction, lighting desired by the devs, and DLSS 5 decides to smooth everything out like an Instagram filter. The examples on Starfield also caused a lot of reactions — completely transformed textures that lose all the patina desired by Bethesda.

PCGamer summed up the general feeling by saying that the writing was "not entirely positive" after seeing the demos. And on the forums, it's even worse. The idea that NVIDIA's proprietary AI could change the rendering of a game without the player having a say, that doesn't sit well.

As much as DLSS 5 will be exclusive to RTX 50 . The RTX 40, probably a lighter version at best. The RTX 30 and earlier? Forget.

Faced with the backlash, Jensen Huang did not hesitate to react. During a Q&A with Tom's Hardware at the GTC, he was direct: They're completely wrong. Players are wrong. There.

Its main argument: DLSS 5 is not a post-processing filter that is slapped onto the final image. It operates at the level of geometry and textures, merging generative AI with the game's 3D data. Developers maintain total control via intensity sliders, color grading, masks to protect certain areas — in short, it's not an Instagram filter on your Resident Evil game.

Todd Howard from Bethesda supported this speech, talking about "incredible possibilities" for future titles. Capcom also seemed enthusiastic behind the scenes.

But well. Telling players they are "completely wrong" is a choice of communication... bold. Especially when your own demos show results that even specialized journalists raise an eyebrow at. Tom's Guide described the technology as a "breakthrough" while noting that if the tuning is poorly done, it can indeed distort a game. The issue may be less with the technology itself than with how NVIDIA presented it — by pushing all the sliders to the max to impress, which had the opposite effect.

Basically, what crystallizes anger is a question that goes beyond DLSS. How far should AI intervene in the rendering of a game? A developer spends months calibrating the lighting, atmosphere, and color palette of their game. If an NVIDIA technology comes along and decides that "no, actually, it would be better like this" — even with control tools — does it respect the original vision?

It's a bit like the same debate as photo upscaling on smartphones. Your Pixel or Galaxy's AI "enhances" your photos automatically. Sometimes it's great. Sometimes, it smooths your grandmother's skin until she's unrecognizable. The question is: who decides what is "better"?

NVIDIA responds: developers. With the right tools. But gamers respond: show us convincing results first, promises after. And for now, the score is not there — at least not with what we saw at the GTC demos.

• Release date — autumn 2026, no specific date
• Compatibility — RTX 50 (5070, 5080, 5090 confirmed), RTX 40 uncertain, RTX 30 and earlier excluded
• Partner studios — Bethesda, Capcom, Ubisoft, Tencent, Warner Bros. Games, NCSOFT, Hotta Studio, S-GAME, NetEase
• Driver required — GeForce 595.79 WHQL minimum (for associated features)
• Framework — integration via Streamline SDK, control tools for devs

Let's be honest: on paper, DLSS 5 is fascinating. The idea of bridging the gap between real-time rendering and cinematic rendering is the Holy Grail of video games for 30 years. And the technical specs — neural reconstruction at the geometric level, not just simple post-processing — are clearly a cut above what the competition is doing.

But. The demos did more harm than good. Showing "improved" characters that look like L'Oréal ads renders, that wasn't the move. And Jensen's response — telling players they're wrong — is the kind of thing that ignites Reddit and Twitter faster than an RTX 5090 benchmark.

What to remember: DLSS 5 will probably be very good when developers integrate it correctly , with subtle settings and not sliders pushed to the max. It's a tool. A powerful tool. But a tool that, if misused, can turn The Witcher 4 into a soulless tech demo. Everything will depend on the implementation — and that, we won't know until the first games this fall.

In the meantime, if you still want to enjoy the best of NVIDIA without bothering with controversial neural rendering, DLSS 4 Multi Frame Gen on the RTX 50 is an absolute killer for performance.

No. It's an option that can be activated or not, like the current DLSS. Developers choose to integrate it, and the player can activate or deactivate it in the graphic settings. If you don't like it, you can turn it off.

Probably not in its full version. NVIDIA has confirmed that DLSS 5 is designed for RTX 50, which have the necessary Tensor Cores to run the neural model in real time. A lighter version for RTX 40 is not excluded, but nothing has been officially announced.

It's more nuanced than that. The demos presented at the GTC were intentionally pushed to the extreme to show the capabilities of the technology, which sometimes led to excessive results. In real-world use, with settings calibrated by developers, the result should be much more subtle. But as long as we don't have final games in front of us, the question remains open.

Not at this level. AMD continues with FSR 4 (upscaling + frame gen) and Intel with XeSS. These technologies improve performance without affecting the intrinsic visual quality of pixels. Neural rendering like DLSS 5 is an approach that only NVIDIA currently offers. This could change, but not for a while.

No, the RTX 5070 and 5080 will also be compatible. However, since the demos were running on two RTX 5090, one may wonder what performance we will have on a 5070. NVIDIA will have a few months to optimize all of this before the release this fall. We recommend waiting for the benchmarks before spending your budget.