Nvidia CEO Jensen Huang says DLSS 5 backlash is 'completely wrong' because it 'doesn't change the artistic control'

Nvidia CEO Jensen Huang has responded to the widespread criticism of DLSS 5. Nvidia's new generation of upscaling technology majors on AI-enhanced lighting but also introduces generative AI at the geometry and asset level and can result in dramatic changes to the look of character models in particular. And that has many observers up in arms, or at least making fun of DLSS 5.

Tom's Hardware attended a Q&A at Nvidia's ongoing GTC event and queried Huang about the negative response to DLSS 5. According to Huang, the critics have it wrong.

"First of all, they're completely wrong," Huang said, "the reason for that is because, as I have explained very carefully, DLSS 5 fuses controllability of geometry and textures and everything about the game with generative AI."

Importantly, he said, developers can "fine-tune the generative AI" and that it "doesn't change the artistic control." Whether that's making a "toon shader" or making the game look like it's "made of glass," DLSS 5 apparently opens up all manner of possibilities, all of which are in the control of the developer.

Of course, the granularity of that control is likely critical here and remains somewhat of an unknown. It's not entirely clear if using DLSS 5 from a developer perspective will be like choosing from a relatively limited menu of "looks" or "styles" or if it's much more open ended.

If the former, well, that's when the fears expressed over the past day or so kick in. The idea that all games with be slathered with the same generative AI gloss is, essentially, what many object to.

But the reality could be both. In other words, there could be some pre-baked settings that developers can choose from for expediency while also having the option of diving in and getting fully involved in how DLSS 5 interprets everything from lighting to materials and assets.

Huang's comments certainly seem to suggest extensive developer control. "All of that is in the control — direct control — of the game developer," Huang said, "this is very different than generative AI, it’s content-control generative AI. That’s why we call it neural rendering."

The other problem with DLSS 5, at this stage at least, is computational intensity. For the demos Nvidia has been showing at GTC, DLSS 5 has been running on a second RTX 5090 graphics card in parallel to a primary 5090 running the game engine itself. That's obviously far from practical or cost effective.

Nvidia says it has DLSS 5 running on a single GPU "in the labs". But if so, why not show the demos running on that single GPU solution? Whatever, DLSS 5 in its current form is clearly a very demanding model and the question remains over whether Nvidia will be able to scale it down into something that can not only run on a single GPU, but can run on the kinds of more affordable GPUs that the vast majority of gamers use. It needs to run on RTX 5060s, not 5090s.

If there is an uncontroversial appeal to DLSS 5, it's the prospect of enabling the lighting realism of ray tracing and path tracing but without the debilitating computational load of calculating all that light bouncing around a scene. To deliver on that proposition, DLSS 5 will need to be far, far more efficient than the dual-5090 hardware setup required for the demo implies. And that improvement needs to come quickly. DLSS 5 will supposedly be released later this year.