DLSS vs Non DLSS

I don't remember when exactly I heard about it but it wasn't today...the info was out there...was it announced the same day as the official DLSS 3.5 announcement- probably not...but the info has been out there...bait and switch to get people to buy a 40 series card?...I'm not sure that would be something that would totally flip the narrative and get people that haven't already bought a 40 GPU to all of a sudden buy one
RR was also review embargoed a day after Phantom Liberty. They were trying to hide it.
 
RR was also review embargoed a day after Phantom Liberty. They were trying to hide it.

I think it all depends on when RR gets released into standard RT...if it's sooner rather than later then all the conspiracy stuff will go away...as far as the RR review embargo, I think Nvidia wanted a big showcase title to debut DLSS 3.5 and it doesn't get much bigger than Cyberpunk
 
And the long, super detailed Digital Foundry video from a day or two ago, where they had a major AI Dev from Nvidia: that video did not say anything about RR only being available for path tracing.

So what I took away from this is ray reconstruction only works for fully path traced/RTX titles - otherwise it's just the new and regular IQ improved DLSS 3.5 for all other DLSS (even for non path-traced but still regular ray-tracing) titles

And expect the exclusive DLSS feature of the 5K gen to be particle/volumetrics AI accelerated rendering/reconstruction like with ray reconstruction for path tracing

Edit: So even in Cyberpunk just FYI - I don't have the game but only Overdrive Mode is path tracing or something? You will only get ray reconstruction only in that mode

It is in the video somewhere cause that's how I found out, as you said no mention of that before this video
 
Last edited by a moderator:
It is in the video somewhere cause that's how I found out, as you said no mention of that before this video
Ok so, at 11:25, Alex mentions that RR had so far only been talked about for games which have a path traced mode. He then asks if RR can be used for games where Ray Tracing is more of a hybrid, added effect (Rather than completely replacing a large portion of the lighting model.). And Bryan Catanzaro says yes it can be used in those hybrid situations. But using it for path tracing is essentially a bigger win (because more visual effects can get adjusted, overall) so they are focused on that.

So I guess the info is kinda there. But that doesn't really tell me that RR will only be available for path tracing mode at launch. At least, not explicitly/clearly.

Its also annoying. Because plenty of non-path traced games with Ray Tracing, implement Ray Traced reflections, ambient occlusion, and Global illumination. All of which are improved by RR. And encompass the bulk of the easily noticed visual effect of Ray Tracing. So, uh....I hope they get to work on that quickly. Because Path Tracing is really difficult to use (performance wise) right now. Whereas plenty of games with RT mixed in, are pretty usable. Especially with DLSS involved.
 
Ok so, at 11:25, Alex mentions that RR had so far only been talked about for games which have a path traced mode. He then asks if RR can be used for games where Ray Tracing is more of a hybrid, added effect (Rather than completely replacing a large portion of the lighting model.). And Bryan Catanzaro says yes it can be used in those hybrid situations. But using it for path tracing is essentially a bigger win (because more visual effects can get adjusted, overall) so they are focused on that.

So I guess the info is kinda there. But that doesn't really tell me that RR will only be available for path tracing mode at launch. At least, not explicitly/clearly.

Its also annoying. Because plenty of non-path traced games with Ray Tracing, implement Ray Traced reflections, ambient occlusion, and Global illumination. All of which are improved by RR. And encompass the bulk of the easily noticed visual effect of Ray Tracing. So, uh....I hope they get to work on that quickly. Because Path Tracing is really difficult to use (performance wise) right now. Whereas plenty of games with RT mixed in, are pretty usable. Especially with DLSS involved.

I agree - which is hilarious because even if you like Nvidia /DLSS/ray tracing etc you can't even deny

NVIDIA👏WHO👏THE👏FUCK👏IS👏IN 👏CHARGE👏OF👏MARKETING👏OVER👏THERE👏WHAT👏THE👏FUCK👏

One Ouija Board after another of marketing material and information
 
Ok so, at 11:25, Alex mentions that RR had so far only been talked about for games which have a path traced mode. He then asks if RR can be used for games where Ray Tracing is more of a hybrid, added effect (Rather than completely replacing a large portion of the lighting model.). And Bryan Catanzaro says yes it can be used in those hybrid situations. But using it for path tracing is essentially a bigger win (because more visual effects can get adjusted, overall) so they are focused on that.

So I guess the info is kinda there. But that doesn't really tell me that RR will only be available for path tracing mode at launch. At least, not explicitly/clearly.

Its also annoying. Because plenty of non-path traced games with Ray Tracing, implement Ray Traced reflections, ambient occlusion, and Global illumination. All of which are improved by RR. And encompass the bulk of the easily noticed visual effect of Ray Tracing. So, uh....I hope they get to work on that quickly. Because Path Tracing is really difficult to use (performance wise) right now. Whereas plenty of games with RT mixed in, are pretty usable. Especially with DLSS involved.

I'm guessing it's just kind of borked for one reason or another in that scenario where there's a handful of separate effects rather than path tracer where everything just falls out of the single algorithm.

I played around with the update for a couple hours. It seems like it's generally a quality win, but occasionally does weird shit. Maybe like 60% of the time it's an improvement, 20% it's neutral, and 20% of the time you seem to lose detail. It seems to touch the entire image in a way I didn't quite expect and has very weird interactions with lower DLSS settings - ultra performance amusingly looks like that Photoshop oil painting filter, it's really, really bad. The image isn't just soft, it's actually horrid. Same with DLSS Auto - seems like it occasionally drops the quality and all of a sudden the image starts looking terrible. Capping at quality seems to be the only way to go at 1440p, but maybe you can get away with lower on a 4k screen.

This is what it looks like. Every so often I guess Auto thinks you're struggling and you start getting this for a moment.
Just capping it at Quality fixed it.

1695356534147.png


I read they also have visual artifacts at native resolution strangely enough, which is why it's disabled.

So I wouldn't be surprised if they just pushed out the first working iteration with some limitations because it misbehaves or produces not so great results otherwise.
 
Last edited:
I'm guessing it's just kind of borked for one reason or another in that scenario where there's a handful of separate effects rather than path tracer where everything just falls out of the single algorithm.

I played around with the update for a couple hours. It seems like it's generally a quality win, but occasionally does weird shit. Maybe like 60% of the time it's an improvement, 20% it's neutral, and 20% of the time you seem to lose detail. It seems to touch the entire image in a way I didn't quite expect and has very weird interactions with lower DLSS settings - ultra performance amusingly looks like that Photoshop oil painting filter, it's really, really bad. The image isn't just soft, it's actually horrid. Same with DLSS Auto - seems like it occasionally drops the quality and all of a sudden the image starts looking terrible. Capping at quality seems to be the only way to go at 1440p, but maybe you can get away with lower on a 4k screen.

This is what it looks like. Every so often I guess Auto thinks you're struggling and you start getting this for a moment.
Just capping it at Quality fixed it.

View attachment 600394

I read they also have visual artifacts at native resolution strangely enough, which is why it's disabled.

So I wouldn't be surprised if they just pushed out the first working iteration with some limitations because it misbehaves or produces not so great results otherwise.
Hahaha. Dang that looks baaad. Reminds me of those mods where they try to AI upscale pre-rendered backgrounds of Playstation games like Final Fantasy 8 and 9, etc.
 
No, but it helps. Difference between the same size at 1080p and 4k is massive.
It doesn't help with post process aliasing. That is the whole point of algorithms like FXAA, SMAA, and TAA. One of the worst side effects of post process aliasing is shimmering, which isn't going to be made better by just increasing the number of pixels rendered. DLSS and DLAA can supplant those other post process anti-aliasing methods while not suffering side effects like fringing and vaseline smear.
 
AI upscaling vs non-AI upscaling

FSR 3.1 still lags behind AI upscalers in some ways, obvious even in HUB’s 4K/Q tests with 40% of the games more than slightly worse than DLSS. What is special about those bad 40%?

  • Lower rendering resolutions: 4K/Quality is most often competitive. 4K/Balanced or 1440p/Quality are usually satisfactory but you can see a bigger gap. 1440/Performance or 1080p/anything have improved a lot from FSR 2.2, but the AI upscalers are still significantly better.
  • Failure Modes: FSR has certain input scenarios it doesn’t handle well. The best FSR games are often those that simply avoid those scenarios. In particular, complex alpha particle effects persist a difficult problem for FSR even with high-quality integrations and low scaling factors.
  • Integration difficulty: FSR’s analytical system requires more work from game engines for special inputs: the reactive mask and the transparency & composition mask. Many games don’t care to produce those, since no other upscaler needs them. They're both optional, a compromise for adoption; FSR makes do without them but at a cost in quality.
In all areas above, AI-based upscalers often perform better because they “hallucinate” around missing or lower-quality information. It turns out AI hallucination is a problem in the general case but not always. Extra detail that’s fake/wrong compared to ground-truth is still great, if:

  1. You don’t have the ground truth! Always true when playing a game.
  2. The result looks nice. “Looking nice” does not require being correct.


View: https://medium.com/@opinali/the-ps5-pro-rdna-4-fsr-4-0-d1bff4bcdca3
 
The result looks nice. “Looking nice” does not require being correct.

That part is really important for people to understand/accept. Realtime graphics are hacks on top of hacks on top of hacks to produce things that look as good as possible. Shadow maps, SSR, bump mapping, deferred lighting, none of these have anything to with the correct way of rendering things, but they are useful hacks that look good and do things fast. Neural network upsampling is the same thing. No, the output of it isn't 100% accurate to if you rendered at the target resolution, but it looks a hell of a lot better than just stretching the image and rendering at a lower resolution allows for more detail or more FPS and so on.
 
I see no difference at 4k.
I actually always activate DLSS with Frame Generation and limit FPS to 120 (monitor max refresh rate). If a game can run faster then it's less taxing on the system, less heat and electricity spent.
 
I see no difference at 4k.
I actually always activate DLSS with Frame Generation and limit FPS to 120 (monitor max refresh rate). If a game can run faster then it's less taxing on the system, less heat and electricity spent.
Should limit 2-3fps below refresh.
 
DLSS framegen automatically activates Nvidia reflex----which does that automatically, among other things.
News to me. I've activted framegen and shot over my max refresh rate without a framerate cap enabled.

I've been capping 2-3 under max refresh since I got a Gsync display in 2016, has never let me down.
 
Back
Top