“Well, first of all, they’re completely wrong,” Huang said in response to a question from Tom’s Hardware editor-in-chief Paul Alcorn about the criticism.
“The reason for that is because, as I have explained very carefully, DLSS 5 fuses controllability of the of geometry and textures and everything about the game with generative AI,” Huang continued.
Just a elongated way to say AI slop.
My main problem is that, even if the technology were to work exactly as advertised, for the first time, PC gaming is moving to an era where our games no longer look the same.
We’ve all seen how every previous version of DLSS worked. The promise of lowering system requirements of high resolutions caught enough attention to draw a crowd, game devs used it to increase the system requirements for lower resolutions instead, AMD and Intel had to develop their own different implementation to stay competitive, and lots of mainstream gaming now requires DLSS and the like on all but the most expensive cards (sometimes even then).
The tech will get better. It always does. Whether that improvement will come fast or slow doesn’t really matter. I think most points of controversy will fade as we’re starting to see this applied to new games. After all, if a game were to be developed knowing that it would be using DLSS 5, the argument that artistic intent will be violated will hold less strength. If DLSS 5 were to ruin the artistic intent of a ground-up DLSS 5-developed game, it wouldn’t be shipped. Future titles being “DLSS 5±mindful” and future improvements to the AI used should get the technology working ever closer to the vision that is being advertised today.
But when that starts to happen, the story will play out exactly the same as it always has for DLSS. It creates results that are “good enough”, so game devs move their development resources elsewhere. Then some management layer catches onto this tech, and finds out that they can just take these resources and move them right into their personal paycheck. End result is games that look generations old and barely run unless you use DLSS 5. By then FSR and XeSS will have to follow suit; if they don’t, AMD and Intel will have the cards that are not only comparatively featureless and weaker, but also produce uglier visuals. But because these generative AI models will never be the exact same between these manufacturers, the output will also never be the same.
And then we move to the weird reality where games look different on different cards, and driver (or, in the case of consoles, system) updates could fundamentally change what games look like, independent of developer input (turns out that artistic integrity can still end up in jeopardy, huh). And unlike with the current AI upscalers we have, where there is a ground truth for the result and we can always obtain it with better hardware, DLSS 5-developed games will always have to rely on it for their “fancy effects” (read: looking like something other than a GameCube game in 2030).
That’s my problem with DLSS 5
Games have had graphics settings for quite some time