Sure, but taking a step backwards in performance and adding lag and then trying to hide it in post processing with what is essentially a motion blur and morphing on streroids isn't advancement in technology. People are calling them out on their bullshit, nothing more.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
Fake frames: An uncomfortable truth about DLSS 5
- Thread starter erek
- Start date
Sure, but taking a step backwards in performance and adding lag and then trying to hide it in post processing with what is essentially a motion blur and morphing on streroids isn't advancement in technology. People are calling them out on their bullshit, nothing more.
But even then raster aside why aren't AMD or Intel developing different technologies than upscale and framegen to again beat Nvidia and show them up? Why have all the top minds/engineers/mathematicians from all these other companies determined this is the way to improve performance?
Hardly. We've hit a brick wall with GPU performance due to manufacturing limits. There's a difference.
Hard to do when nVidia controls so much of the market. Their only hope is to beat them in raster first before attempting anything else.
- Joined
- Dec 19, 2005
- Messages
- 17,421
agree
I am not an engineer and all this is beyond my understanding. Maybe we are reaching the end of how computer graphics are being generated, maybe things need a complete overhaul and rethink new ways, maybe voxels can make a comeback and take things further, or maybe not, I do not know and it is not my job to know. What I do know is that frame gen as it is now is not the way.
That's still hitting a limit on raster one way or another - distinctions without a difference. Like saying the reason you can't cross a busy street isn't because you're blind, it's because you have no legs
Why aren't they doing that then?
I don't see how you can say you know it's not the way when you admit you don't know the way. Those two thoughts contradict each other.
"Hitting limits of raster" is a Jensen talking point to push AI since that's where their money is coming from right now. 40 series was a huge leap over 30 series in raster performance and was not that long ago. 50 series was much smaller, but a 30 or so % improvement given it's the same process node isn't atrocious, the pricing is.
- Joined
- Aug 29, 2025
- Messages
- 239
Lmao I'm loving these
This tech will surely be a boon to gooners
This tech will surely be a boon to gooners
So then again why isn't AMD or Intel just pumping out + beating Nvidia then on pure raster alone cards? Wouldn't that be a good talking point for them to have, and then not just be a talking point, but a selling point to then show the card that can do so in their hands and sell it to people like you who want it? Why hasn't that happened and seemingly not going to?
Was it though? 50% jump on the XX90 & XX80 levels, but diminishing down the stack until, at the always most prevalent XX60 level, the improvement was minimal.
Because I am talking about two different things here, for petes sake. Actually designing a GPU or figuring out to do computer graphics from scratch vs a simple special effect to artificially make motion look smoother than it actually is, at the cost of latency, an ages old trick that has existed in televisions for decades. What you are saying is that I cannot paint a house without knowing how to build one.
No I'm saying you can't tell me I built my house wrong if you don't know how to build a house yourself.
Video itself is just moving images - the same way computer graphics are creating images that can also optionally move (vs still CGI images) - the technologies are not as unrelated as you try to make them seem.
Gamers have been bought in on the AI junk as anyone else.
Sorry the mass of gamers buying Nvidia cards have been happy they have DLSS, RT and the related gpu compute things. Its been their reason for ignoring faster raster cards at their price point for a few years now. To the point where AMD said fine we'll make a 70 class card and nothing higher, why bother.
Nvidia made a choice with Volta... which they never even released a consumer version of. That they were going to dedicate a large chunk of every die to pure compute bits. They then got to work finding gaming uses for it.
I am not saying DLSS and Framegen don't have their place. Lets not pretend though that Nvidia could not have chosen instead to ramp raster performacne up. Stagnating Raster performance (mostly) was a choice. They have lifted raster each gen they just have choosen to target 10-20%. At any point they could have made the choice to include more ROPS at the expense of a couple RT/compute cores.
Sorry the mass of gamers buying Nvidia cards have been happy they have DLSS, RT and the related gpu compute things. Its been their reason for ignoring faster raster cards at their price point for a few years now. To the point where AMD said fine we'll make a 70 class card and nothing higher, why bother.
Nvidia made a choice with Volta... which they never even released a consumer version of. That they were going to dedicate a large chunk of every die to pure compute bits. They then got to work finding gaming uses for it.
I am not saying DLSS and Framegen don't have their place. Lets not pretend though that Nvidia could not have chosen instead to ramp raster performacne up. Stagnating Raster performance (mostly) was a choice. They have lifted raster each gen they just have choosen to target 10-20%. At any point they could have made the choice to include more ROPS at the expense of a couple RT/compute cores.
But they are different. Frame generation is not rendering anything, it is interpolating between two actual frames and guessing what it should look like based on their differences. Those interpolated frames also do not react to your inputs, it has to wait until the next real frame is being drawn and that is being influenced by your inputs.
Interpolation is great for many things video related but not for things that require real time user input. They are always late.
M76
[H]F Junkie
- Joined
- Jun 12, 2012
- Messages
- 15,181
Full 10+ billion parameter models already give that uncanny AI face look, the distilled models look about 10 times worse. I can't imagine how bad the model can be that runs real time on a single gpu.
No, actually I can. Just as I can tell that the food is spoiled without being able to cook.
You appear to think I was implying it's a trivial matter, it isn't. AMD and Intel would need to design an architecture capable of doing that which would involve a lot of R&D which costs a lot of money, not sure if you're aware, their market share is virtually nothing so the odds they'll be able to recoup that investment is next to nil. It's not economically feasible for them to compete with nVidia at that level.
Yes it was, there was no 4090Ti, which would be the logical successor to the 3090Ti, but even without one, the 4090 appears to be on average 42% faster based on those numbers. If we compare it it's immediate predecessor, the 3090 its a 60% improvement. That's pretty significant to me, but even if it's not to you, we clearly aren't at the "limit of raster" if we were those numbers would be much closer.
The last time AMD/anyone could make a HALO to best Nvidia's HALO, at raster at that, was over a decade ago/13 years ago - seems the market/tech leader (and all their mathematicians and engineers) knows what they're doing, especially if they felt they had to pivot from raster themselves, when still no one could beat them.
Edit: And before anyone jumps in - 6900 XT only sometimes beat the 3090, this ain't horse shoes as they say - sometimes and almost doesn't count.
Edit: And before anyone jumps in - 6900 XT only sometimes beat the 3090, this ain't horse shoes as they say - sometimes and almost doesn't count.
Last edited:
Pivoting from raster is an economic decision as well. The AI space is still very green and advancements are happening quickly. DLSS 4.5 was literally just the other day and here we are talking about DLSS 5. Not only is it advancing quickly, but they're able to apply it to tech they've already developed.
Like the other saying goes - when God/whoever closes a door (raster) a window (AI) opens.
If raster was so easy to achieve + all this other AI tech was so hated as some try to make it seem, again AMD/Intel/someone else would serve that market and pump out a raster only card that beat Nvidia with all their non-raster tech. And again it would apparently be gobbled up by the people, we're led to believe. Yet the former can't seem to be achieved, and even when they try to get as close as they can, yet still fail, the latter doesn't happen either, again as with the 6900 XT (but muh Mindfactory!!!!!)
Raster isn't easy anymore, because Nvidia has gamers believing they need acronym technologies to play games, to the point games shipping now are forcing RT and upscaling.
There isn't any limit to raster, that's simply Jensen marketing people on not increasing raw performance. AMD or Intel could make a larger powerful raster card, but why would they? Manufacturing a large die is expensive, there is no point when they can just cut costs with a cheaper die and push upscaling.
Unsurprisingly the next step is AI slopifying games ruining the original game art for generative AI being thrown on top of the original image.
There isn't any limit to raster, that's simply Jensen marketing people on not increasing raw performance. AMD or Intel could make a larger powerful raster card, but why would they? Manufacturing a large die is expensive, there is no point when they can just cut costs with a cheaper die and push upscaling.
Unsurprisingly the next step is AI slopifying games ruining the original game art for generative AI being thrown on top of the original image.
The difference is that with more ROPs you get faster raster as long as chip manufacturing can keep doubling size then it will scale. RT OTOH requires tons of tricks and low fidelity gimmicks in order to do because the computational cost is exponentially more. After you implement ray reconstruction, AI upscaling, AI frame gen, are you even looking at more accurate lighting or just a rendering pipeline that one company can control? That's a brick wall, with a gate. Raster? Not at all.
Because the only way nVidia pulls it off is using it's marketshare to justify the larger dies. Even if AMD or Intel were to produce a card with faster raster it would take eons and tons of cash to get gamers to switch.
I mean, there literally is a limit to raster.
Certain effects are basically impossible or don't scale well after a point. This is the idea behind megalights in Unreal for example. There's a breakpoint in there somewhere, where once you want a certain amount of certain things on screen, ray tracing starts scaling _better_.
Once you get past the ray tracing performance task, getting more lights and shadows on screen is extremely cheap. Want 300 shadow casting drones shining light everywhere? Suicide in raster, but barely matters to a ray tracer.
Certain effects are basically impossible or don't scale well after a point. This is the idea behind megalights in Unreal for example. There's a breakpoint in there somewhere, where once you want a certain amount of certain things on screen, ray tracing starts scaling _better_.
Once you get past the ray tracing performance task, getting more lights and shadows on screen is extremely cheap. Want 300 shadow casting drones shining light everywhere? Suicide in raster, but barely matters to a ray tracer.
As opposed to using performance decreasing tricks and gimmics like Ambient Occlusion? What's better chocolate or vanilla? I'd say the trick that aims to more accurately mimic actual physics and actual reality, as early on as it is. And I'm just talking about ray tracing here. Wait till you find out all the tricks your eyes and brain do to 'actually see the real world' then we could even possibly include 'perceptual' tricks into that mix also DLSS 5 etc
Marketshare they got by making the better tech. Goodwill and marketshare Intel and AMD lost by not (thus would cost more to recapture).
Who said easy? I literally replied to you an hour ago acknowledging it’s not a trivial matter the first time you implied I said it was easy.
You were pointing out it's an economic decision - I'm pointing out because you hit a wall/a door closed does not mean the open window - even if an economic decision - is still not the way - even if it weren't an economic decision (but even then Nvidia wants to sell more cards - even to gamers - so whatever their course was/would be would also be an economic decision - even if the current course of AI never existed)
It is an economic decision and raster has not hit a wall. None of that translates to easy.
Doesn't translate to bad or wrong either.
Bad, wrong and easy are all words you’re imagining exist in my post. The first two being pretty subjective metrics.
Nvidia is a company, everything they do is literally an economic decision - so what does pointing out it's an economic decision show or supposed to mean by you then? Tell us, in your own words then.
You're preaching to the choir in terms of what a human's eyes and brain can recognize. You think the human brain can calculate an accurate ray in real time? The human brain is horrible at it which makes the need for ray tracing all that more comical. You have people telling you things are accurate when in fact they aren't. Alan Wake 2 for instance great game, loved it but good god is the sun blown out in most scenes with some questionable accuracy in some places that even novices were able to point it out.
Is ray tracing better tech? Is AI? Is this truly better or just completely different? Is it higher fidelity or a mod? Are they selling a problem and then the solution?
I'll leave the rest up to the reader's opinion/buyer's wallet, but I'd argue yes on ray tracing again:
What is AMD selling (or not, but trying to), charity?
If the human brain is horrible with determining appropriate lighting I would say it's not worth the cost.
AMD is copying nVidia which I firmly disagree with. Its not like if they do RT I all of a sudden like it.
Well then I guess you'll either just have to get dragged kicking and screaming into the future or stop buying GPUs of any brand and any games is the most you can do instead
Everything I’ve said has been my own words. Feel free to use your scrolling functions.
Right so what did you mean/what were you pointing out when you said:
in your own words? Feel free to be able to explain your own words. Or not.
Forcing upscaling, and now AI hallucinated lighting instead of pre-baked in lighting are all economic decisions that have already been explained in this thread, it's about cutting costs on the gaming division while Nvidia is raking record profits on AI.
It's clear they're caring less and less about their bread & butter market that has built up Nvidia to where they are now.
I don't like that AMD is copying Nvidia, they're damned it they do, and if they don't. I guess we'll see if they go along with the AI image filter effect.
It's clear they're caring less and less about their bread & butter market that has built up Nvidia to where they are now.
I don't like that AMD is copying Nvidia, they're damned it they do, and if they don't. I guess we'll see if they go along with the AI image filter effect.
ROI for prioritizing AI based tech > ROI for prioritizing pure raster performance.
I already alluded to this earlier. Interesting you didn’t pick up on it but instead picked up on things I never said or alluded to, twice.
If gamers don't bring in the majority of money/profits anymore - and AI does - then gamers are literally not the bread and butter by definition. But regardless - Nvidia still brought in $3.6 billion from gaming in their last quarterly report (vs AMD bringing in just $843 million from gaming in comparison - Intel doesn't even disclose it's that low - thrown in with 'Client Computing' in their reports)
If anyone thinks Nvidia - or Jensen - is going to willingly give up that money on the table - or cede this/these crowns to anyone else (as much as you wish he would):
then I also have a bridge in Brooklyn to sell you and you should know the people you're hating better.
In your buddy's own words:
To which:
- Joined
- Dec 19, 2005
- Messages
- 17,421