NVIDIA discussion [2024]

  • Thread starter Deleted member 2197
  • Start date
“This trend, however, worries some gamers who fear the technology will become essential to good performance.”

This is silly. As long as upscaling produces good results why should anyone fear it. There are tons of hacks required for good performance. Upscaling (“hallucinating pixels”) is fine if it actually works.
 
I guess it could be a fear of missing out of sorts. Of feeling forced to upgrade more frequently just to stay updated with the technologies.
Not saying this is justified or un-justified, just feels like someone could plausibly be thinking
 
1 pixel to 32 pixels?

That's way more than the 8 pixels of DLSS performance plus frame gen, or even the 27 of ultra performance plus frame gen.

I wonder if we can infer anything from this...

Perhaps a new ultra performance mode with 25% on each axis?

Framegen ++
 
wonder if we can infer anything from this...
Nice catch!

I am trying not to read much into this, but Jensen has repeatedly boasted before that 1 in every 8 pixels is rendered and the rest are AI generated (talking about DLSS Performance + Frame Generation).

Switching to 1 in every 32 pixels is a huge jump not explained by any current DLSS tech, even if frame gen has double the output, that merely means 1 in every 16, frame gen would have to be quadrupled twice to reach 1 in every 32!

Maybe he is including in Ray Reconstruction? Still doesn't explain the insane jump to 32!

DLSS4 is about to be revealed, I think this strongly hints about what's coming!
 
Last edited:
DLSS ultra performance + 4x frame gen gets you there.

DLSS UP on its own would be 9, then with current FG we double to 18. If we get triple FG we get to 27 and then obviously 36 with quad FG so it doesn't quite fit in with the specific 32 number.

Of course he could have been being deliberately imprecise to obscure the underlying functionality so this doesn't necessarily mean anything, but the way I see it there are only 2 exact ways of getting to 32 which is either a new ultra perf mode at 25% (rather than 33%) on each axis with regular FG, or the existing DLSS P with 8x FG.

The latter doesn't seem technically feasible to me, but the former would be a fairly natural progression of what we currently have if we assume the quality of DLSS is going to jump up this generation allowing for lower input res at a similar output quality.

While a new Ultra Perf mode might not be of huge interest to most enthusiasts, what if an associated quality jump to allow it flows into the higher quality modes as well? E.g. quality is now 50% on each axis, performance is 33% etc...

Highly speculative and wishful thinking, but it would be pretty awesome if correct.
 
His comments don’t necessarily have to be referring to games.

Yeah I am probably reading way too much into this. But the more I think about it, the more it makes sense.

DLSS Super Resolution has for a long time now been THE major differentiator for Nvidia (alongside RT performance). Frame Gen is great but everyone has that now and it's still seen as somewhat of a compromise as opposed to a straight up win anyway due to the latency implications and memory footprint.

However with the launch of this next generation of GPU's, the competition have all caught up with DLSS (yeah DLSS is likely still a bit better but we're playing in the margins now), Intel has XeSS, Sony has PSSR and we know AMD will be following shortly with AI driven FSR4.

I imagine Nvidia is extremely keen not to let the competition catch up in that regard, and 3x frame gen really isn't going to cut it IMO as a major differentiator given the associated compromises.

However if Nvidia could bring out a new iteration of DLSS that builds on it's quality lead to such an extent that it can match the other AI based upscalers at a significantly lower resolution (lets say 1080p vs 1440p) then that would be a HUGE differentiator and essentially maintain the lead that DLSS has given them over the past 6 years. Given Nvidia's position in the AI market and their already healthy quality and time lead with DLSS, if anyone can pull this off, it's them.
 
His comments don’t necessarily have to be referring to games.
Was going to say this. What Nvidia does and can do internally doesn't mean they're at that point in consumer hardware. Though I'm absolutely sure that Bryan and the DLSS team at Nvidia are cooking up a DLSS3 successor which does more with fewer pixels.

Shouldn't be too long now until we see the next gen of AI upscaling.
 
Back
Top