Vya DomusWhat a bizarre choice to make a point.

It's a driver level feature, how could it be open source lol.

No, that's what Nvidia claims, there is no way to know that for sure because it's not... open source. They've made claims like that which proved not to be true before, like saying that frame generation cannot work on GPUs other than 4000 series because the optical flow whatever was too slow without dedicated hardware and that turned out to be a straight up lie because AMD got it to work on compute shaders which run on everything.

So suffice to say I for one do not believe that.

Oh yes true my bad ! Anti-lag is driver-side, after reading about it, it is certainly going to end up game-side tho :p, it was reportedly deactivated because it worked by injecting code in the games directly and caused ban waves in CS2...I sure wished something that inject codes in my applications would be open-source AND NOT in a driver tho, rather in the game, my mistake.

We'll see once the feature is up again, if it's up, but I do not think it will stay driver-side, just my humble opinion.

Regarding DLSS claims or FG:
AMD FSR3 on Avatar is the best implementation and is very impressive, FG included, but on what basis does that mean Nvidia implementation doesn't actually need the Optic flow generator for DLSS FG ?

You say AMD made FG work, sure and Avatar is impressive, but frame generation has been in most TV for years, something called "soap-opera" effect yet it's not AMD or Nvidia implementation.
Why would AMD managing it differently would imply Nvidia doesn't really need a hardware for THEIR frame gen tech ? You have no evidence of this, only that AMD manages it without optic flow generators

Nvidia's bet is on AI, AI improves over times and using a dedicated AI hardware architecture like in Apple silicon, Nvidia GPU or newest Qualcomm SoC is nothing shocking


Just saying 'I don't believe Nvidia really needs the AI hardware because AMD solution does not' is really too simplistic and not really an evidence for me

sephiroth117Oh yes true my bad ! Anti-lag is driver-side, after reading about it, it is certainly going to end up game-side tho :p, it was reportedly deactivated because it worked by injecting code in the games directly and caused ban waves in CS2...I sure wished something that inject codes in my applications would be open-source AND NOT in a driver tho, rather in the game, my mistake.

We'll see once the feature is up again, if it's up, but I do not think it will stay driver-side, just my humble opinion.

Regarding DLSS claims or FG:
AMD FSR3 on Avatar is the best implementation and is very impressive, FG included, but on what basis does that mean Nvidia implementation doesn't actually need the Optic flow generator for DLSS FG ?

You say AMD made FG work, sure and Avatar is impressive, but frame generation has been in most TV for years, something called "soap-opera" effect yet it's not AMD or Nvidia implementation.
Why would AMD managing it differently would imply Nvidia doesn't really need a hardware for THEIR frame gen tech ? You have no evidence of this, only that AMD manages it without optic flow generators

Nvidia's bet is on AI, AI improves over times and using a dedicated AI hardware architecture like in Apple silicon, Nvidia GPU or newest Qualcomm SoC is nothing shocking


Just saying 'I don't believe Nvidia really needs the AI hardware because AMD solution does not' is really too simplistic and not really an evidence for me

Sure nvidia solution requires OFA, because it was made that way. Even better is the fact that RTX 3xxx and 2xxx already has OFA, just not as performant as in 4xxx series.
That does not change the fact that if they wanted they could made it to work with older cards. The decision to not do it was purely a business one - to sell more 4xxx cards as without FG buying 4070 or something other than 4090 makes even less sense.
Whole marketing from nvidia was based of giving the performance numbers of those cards with DLSS3 FG enabled vs 3090 without it. If 3090 would be able to do it those graphs would not look that impressive.

fancuckerI'd rather pay more for a proper hardware based and closed solution like DLSS. The vast majority of consumers believe in subsidizing Nvidia's attempts at advancing graphics and gaming.

Nothing is stopping you from paying more.

And proper hardware?? Isn't DLSS, FSR, XESS, software layers, that tell the hardware what to do?

I'd rather just have a card that can run the monitor's default refresh rate & resolution...................and I'd pay more for it ;)

sephiroth117You say AMD made FG work, sure and Avatar is impressive, but frame generation has been in most TV for years, something called "soap-opera" effect yet it's not AMD or Nvidia implementation.

You don't understand that this makes your point even weaker, this makes Nvidia's lies even worse. You're not correct anyway, it's one thing to get this working in a TV where latency is irrelevant and a completely different thing to make it work in software on a PC that's rendering something in real time, if you have seen it in TVs you know that it looks quite bad at times.

sephiroth117Why would AMD managing it differently would imply Nvidia doesn't really need a hardware for THEIR frame gen tech ? You have no evidence of this, only that AMD manages it without optic flow generators

You don't understand how this works, AMD is using an optic flow generator as well, it's just that theirs works with compute shaders on everything and is plenty fast so it doesn't need hardware acceleration.

I don't know why it's hard for you to accept that Nvidia simply lied on this one, they claimed the hardware acceleration was necessary, which it wasn't. FSR FG works on Nvidia GPUs as well in case you didn't know and most people agree it's basically the same.

sephiroth117Just saying 'I don't believe Nvidia really needs the AI hardware because AMD solution does not' is really too simplistic and not really an evidence for me

There is plenty of evidence, Intel also has an "AI upscaler" that works on everything as well, strange how only Nvidia's implementation needs their own unique hardware for it to work isn't it ? lol

nguyenI guess Apple and MS got rich and popular because they keep making open source softwares then :roll:

Open or closed, if it gets the job done and comes at a reasonable asking price... i'm game!

One thing i definitely don't do is look at the Fortune Global, Forbes, Bloombergs index, etc to determine where to inject my hard earned cash :p

I do like Apple though... healthy fruit

AssimilatorA 4080 isn't NVIDIA's most powerful GPU. Please learn reading comprehension.

I don't get it. The topic here is FSR 3, which has nothing to do with which is the most powerful GPU. It is factual that Nvidia have more powerful hardware (which is not cheap by the way), and better software features. However for the latter, they tend to leave their own customers behind after a number of years. Can an Ampere GPU run Nvidia Frame Gen? No, but yes for FSR 3.0 Frame Gen. So honestly, I don't see why people keeps fending for Nvidia. As an Ampere GPU user, I am also annoyed by Nvidia for deliberately shutting out features on me.

noel_fs...also the drivers have become pretty decent the last few months, only took them 2 years, could have been worse, or could it?

Always! And it only took them 22 years, not 2. These open sauce things are mostly for perception.

Honestly the Nvidia butt plugging in this threads disgraceful, all the usual trolls out and In force trying to derail a thread into yet another battle, truly schill like , shareholder behaviour.

watzupkenI don't get it. The topic here is FSR 3, which has nothing to do with which is the most powerful GPU. It is factual that Nvidia have more powerful hardware (which is not cheap by the way), and better software features. However for the latter, they tend to leave their own customers behind after a number of years. Can an Ampere GPU run Nvidia Frame Gen? No, but yes for FSR 3.0 Frame Gen. So honestly, I don't see why people keeps fending for Nvidia. As an Ampere GPU user, I am also annoyed by Nvidia for deliberately shutting out features on me.

Would be nicer of AMD to ask their sponsored titles to include Nvidia Reflex, so Nvidia owners can get better gaming experience running FSR3 than Radeon :D

sephiroth117Why would AMD managing it differently would imply Nvidia doesn't really need a hardware for THEIR frame gen tech ? You have no evidence of this, only that AMD manages it without optic flow generators

This, AMD has the benefit of being able to learn from it already being done, to try and find other ways to achieve the same thing (you can bet over at AMD the MO was study Nvidia FG and find the quickest way to make an acceptable hackjob alternative). Because it can be done now, years later and achieve acceptable results doesn't necessitate that Nvidia lied about needing the Ada OFA to do FG the way it's currently done/developed for Ada.

This is especially true because it's evident that like Upscaling, each camps unique approach differs in R&D, execution and quality (and perhaps more), to just say these two companies 'did the same thing, so one is a liar', is far too vague and accounts for none of the many differences and nuances between them from conception to completion.

sephiroth117Just saying 'I don't believe Nvidia really needs the AI hardware because AMD solution does not' is really too simplistic and not really an evidence for me

Also spot on, not nearly enough evidence to draw that as a definitive conclusion, given the above factors especially. XeSS as an example thrown out, sure it has a compatibility code path that works across vendors.... that delivers unequivocally worse results both in performance and image quality. Side note but it's impressive that XeSS (v1.2) has already been able to produce better IQ than FSR 2.2.x at equal input resolutions, and subjectively is still often preferable when performance normalised.

I'm very keen to see AMD's FG continue to grow and be implemented, and to test it out for myself. I have no real issue with the 'innovation but locked', followed by 'copied (hackjob) and open' that we're getting in cycles lately. I'd certainly prefer that over no innovation.

I dunno, maybe they open sourced it because that have custom customers that run their own OSes, like Steam Deck and PlayStation?

wolfTypically what I've seen is you don't get the open one without the closed one first. Sure it's better when it eventually comes out and catches up, but I don't think we'd have gotten FSR (upscaling or FG) without DLSS trailblazing it. Chicken and egg stuff.

Nvidia tends to innovate new features, lock them to their hardware and get a generation or two of exclusivity, then AMD benefits from following that and developing a decent open source version.

I don't really mind the cycle tbh, I am a big fan of innovation and new features, I don't want to play games with the same rendering tech ad infinitum, and it gives good options for those who either want to be at the bleeding edge and pay for it, or wait if they choose.

nguyenI guess Apple and MS got rich and popular because they keep making open source softwares then :roll:

Just a few days later there is experiment going on for a mod putting DLSS data into FSR3 and enable frame gen for Non-RTX 4000 Nvidia GPUs.

github.com/Nukem9/dlssg-to-fsr3

CrackongJust a few days later there is experiment going on for a mod putting DLSS data into FSR3 and enable frame gen for Non-RTX 4000 Nvidia GPUs.

github.com/Nukem9/dlssg-to-fsr3

Looking at the difference in visual quality, rtx4000 owners still have the advantages with FG, just like RTX owners in general have access to the superior DLSS

nguyen

Looking at the difference in visual quality, rtx4000 owners still have the advantages with FG, just like RTX owners in general have access to the superior DLSS

Yeah but it is a mod and it is just a few days on their first release.

Just to show you it is the open solution breaking the artificial barriers set by Nvidia.

nguyenLooking at the difference in visual quality, rtx4000 owners still have the advantages with FG, just like RTX owners in general have access to the superior DLSS

Has promise but from that comparison the ghosting looks terrible, still keen to try it myself and see some trusted sites do comparisons.

CrackongJust to show you it is the open solution breaking the artificial barriers set by Nvidia.

It's an open solution broadly completing the same task, a different way, with it's own drawbacks and nuances, a year or so later. I wonder if it would exist if not innovated by others earlier...

And good on them, good for everyone! AMD have done well to get it working to this level so quickly.

wolfHas promise but from that comparison the ghosting looks terrible, still keen to try it myself and see some trusted sites do comparisons.

It's an open solution broadly completing the same task, a different way, with it's own drawbacks and nuances, a year or so later. I wonder if it would exist if not innovated by others earlier...

And good on them, good for everyone! AMD have done well to get it working to this level so quickly.

Yeah but like mentioned before, it's just a mod that uses nvidia Streamline to basically inject FSR3 code and trick it to think it's DLSS3 FG as far as I understand, not tuned. Anyway it's still just a mod not official implementation.
In Avatar where it's implemented officially it looks very good. Even when I ran it at native 4K with unobtainum settings so base framerate was like 20-30, apart from obvious latency image itself showed almost no artifacts.
Funny considering nvidia whitepapers saying that you need AI to interpolate shadows and particles :roll:
Even funnier is that if their OFA is so good without motion vectors then I guess they will have driver option as well for it as AMD does with driver preview now and soon official driver.

pavleAlways! And it only took them 22 years, not 2. These open sauce things are mostly for perception.

GCN drivers were quite good up until they refused to fix the UVD bug for the early generations, polaris drivers were acceptable at launch just lacking a bit in performance which after a couple years they were really good. RDNA2 was unacceptable and they took almost years to get them decent. Thats 10+ years of history and basically just the last 2 years were bad. Pre-GCN they were fine as far i could tell and i used them well past launch dates.


Has little to do with perception, if they wanted "perception" they would do marketing to brainwash people.

sephiroth117Nvidia has and participates in countless open-source projects

Name 3.

Also, PM me when they actually start caring about their "open" linux driver.

AMD actually has a driver that interfaces with open components well. NVIDIA does not and has actively resisted this. End of story in my eyes.

AssimilatorA 4080 isn't NVIDIA's most powerful GPU. Please learn reading comprehension.

Well, and 7900 XTX isnt AMD's most powerful. Please stop typing misinformation and develop some critical thinking. The 4090 was made commercially avaliable just so that nvidia could say exactly that, "most powerful gaming card", when in reality most of them are going into AI because they are not affordable for 99% of gamers and they dont even perform correctly in most games cause the thing is huge and the gaming drivers werent developed for it, anyway, as a side effect they are cutting down production of the actual consumer cards because AI is more profitable, making all of their cards more expensive just so you have the false belief that nvidia has the most powerful card which you cant even afford.

You need to realize nvidia isnt doing you any favors, thair profit is going into more marketing so that you see their pamphlets everywhere and if you see them everywhere it must be true.

AssimilatorA 4080 isn't NVIDIA's most powerful GPU. Please learn reading comprehension.

Which, tbf, in turn makes the assumption that the 7900XTX was to be a direct 4090 competitor or, worse, a 4090 killer.

I don't recall anybody official or otherwise saying that on behalf of AMD, nor anybody on Nvidia's side officially stating the opposite for the 4090 for that matter.
Nobody really knew until first the launch announcements and secondly, the NDA's were lifted.

I mean, I get it... the 4090 is the best card bar none at the £1500 level for any old 'who, again?' brand or the closer to and over £2000 level for the known ones. But against well known brand 7900XTX's dropping as low as £850 or so and still undercutting 4080's by some £500+ at that point... do they really need to be 1:1 comparisons to be good for x or y?

noel_fsone of the reasons im an amd guy :love:

also the drivers have become pretty decent the last few months, only took them 2 years, could have been worse, or could it?

Tbh they've been good for years at this point, not merely months.

R-T-BName 3.

Also, PM me when they actually start caring about their "open" linux driver.

AMD actually has a driver that interfaces with open components well. NVIDIA does not and has actively resisted this. End of story in my eyes.

There are a lot more than three here.

noel_fsWell, and 7900 XTX isnt AMD's most powerful.

LMAO WTAF, are you fanboys so delusional you're just going to make things up because you know you have no counter-argument?

noel_fsand they dont even perform correctly in most games cause the thing is huge and the gaming drivers werent developed for it

More made-up nonsense.

noel_fsjust so you have the false belief that nvidia has the most powerful card

It's not a false belief. It's objective fact. Something you apparently have trouble with.

CMP01do they really need to be 1:1 comparisons to be good for x or y?

Irrelevant to the claim that was made and the facts I presented to refute it. Stop trying to move the goalposts.

AssimilatorThere are a lot more than three here.

To be clear, I know they have projects and a lot of them, but my point was "from the top of your head, name 3." The thing I was trying to establish is AMD is more involved and has more of an "Open" rep than nvidia.

That being said, that's not really that awful. Just a different way of doing things. Fact is nvidia software products are high quality and that affects my purchase choice a lot more.

AMD is good in open source land, but honestly, the windows driver still has a few issues I'd like dealt with. That's IMO a worse issue.

R-T-BTo be clear, I know they have projects and a lot of them, but my point was "from the top of your head, name 3." The thing I was trying to establish is AMD is more involved and has more of an "Open" rep than nvidia.

That being said, that's not really that awful. Just a different way of doing things. Fact is nvidia software products are high quality and that affects my purchase choice a lot more.

AMD is good in open source land, but honestly, the windows driver still has a few issues I'd like dealt with. That's IMO a worse issue.

They do have a more open reputation, I never said Nvidia was more open, my point was that it's not like Nvidia is doing nothing in the open-source area, especially around IA/ML.

I am not familiar with all of them but I do use two of those open source frameworks, a framework named Apache Spark (through RAPIDS) for distributed processing and PyTorch (machine learning) for work, both have a lot of Nvidia contribution to them and they work wonderfully well with Nvidia GPUs

We don't need to use proprietary nvidia tool to perform ML or data ETLs, they just really did their homework for AI/ML/data scientists on many open-source projects.

For consumers, I think AMD has the upper hand in open-source of course, but right now DLSS3.5 is what I prefer.



I don't care about closed or open upscalers frankly because I feel that many studios want a little paycheck from either AMD or Nvidia to implement it anyways and are not really waiting for things to go open-source, at least in AAA (gpu sponsorship is omnipresent in AA/AAA pc gaming).

I'd even tell you that DLSS SDK that you use to implement DLSS in your game is open source, the algorithm itself is not but anyone can implements it using an open-source, transparent SDK in their code, the rest is in the driver and hardware but at least what you embark in your executable is known:
github.com/NVIDIA/DLSS


I care about quality, implementation rate in games.. DLSS is in far more games, overall it's a really great upscaler relatively speaking, it does not include vulnerabilities or anti-cheat false positives, it's continuously improved with stable features but also experimental ones like ray reconstruction.

FSR3, I can use it on Nvidia whilst DLSS can't be used on AMD, again, despite what some people say here, nvidia upscalers are HARDWARE based, they made that choice, it's AI, it's a neural network that improves through time to a pace a deterministic algorithm like FSR may not be able to match at one point.

Having frame generation locked behind a hardware optical flow generator sucks for < RTX 4000, absolutely, they could have worked at least on a software-based rollback, but saying in above comments that it's all a lie that hardware architecture is useless is naive and baseless, this architecture whole point is to be trained through time by Nvidia to continuously improve the neural network behind DLSS and FG, something a software solution cannot do as well .

sephiroth117They do have a more open reputation, I never said Nvidia was more open, my point was that it's not like Nvidia is doing nothing in the open-source area, especially around IA/ML.

I am not familiar with all of them but I do use two of those open source frameworks, a framework named Apache Spark (through RAPIDS) for distributed processing and PyTorch (machine learning) for work, both have a lot of Nvidia contribution to them and they work wonderfully well with Nvidia GPUs

We don't need to use proprietary nvidia tool to perform ML or data ETLs, they just really did their homework for AI/ML/data scientists on many open-source projects.

For consumers, I think AMD has the upper hand in open-source of course, but right now DLSS3.5 is what I prefer.



I don't care about closed or open upscalers frankly because I feel that many studios want a little paycheck from either AMD or Nvidia to implement it anyways and are not really waiting for things to go open-source, at least in AAA (gpu sponsorship is omnipresent in AA/AAA pc gaming).

I'd even tell you that DLSS SDK that you use to implement DLSS in your game is open source, the algorithm itself is not but anyone can implements it using an open-source, transparent SDK in their code, the rest is in the driver and hardware but at least what you embark in your executable is known:
github.com/NVIDIA/DLSS


I care about quality, implementation rate in games.. DLSS is in far more games, overall it's a really great upscaler relatively speaking, it does not include vulnerabilities or anti-cheat false positives, it's continuously improved with stable features but also experimental ones like ray reconstruction.

FSR3, I can use it on Nvidia whilst DLSS can't be used on AMD, again, despite what some people say here, nvidia upscalers are HARDWARE based, they made that choice, it's AI, it's a neural network that improves through time to a pace a deterministic algorithm like FSR may not be able to match at one point.

Having frame generation locked behind a hardware optical flow generator sucks for < RTX 4000, absolutely, they could have worked at least on a software-based rollback, but saying in above comments that it's all a lie that hardware architecture is useless is naive and baseless, this architecture whole point is to be trained through time by Nvidia to continuously improve the neural network behind DLSS and FG, something a software solution cannot do as well .

I think its great we have 2 different approaches to the same problem (or upscaled solutions), both with pros and cons. The battle of the ages for GPU innovations... yes pls!!

Although Nvidia may have the advantage, dedicated tensor cores comes at a cost, a cost which may be less rewarding in the long run if deployed hardware features fall short of compatibility with later DLSS iterations. Full respect to Nvidia though, it takes more sweat for hardware based solutions to pay off and so far they're championing that race. I'm usually more concerned with price to performance native perf hence don't really get into all the upscaler skirmishes but its great to see innovations from Nvidia/AMD (...Intel) opening new doors on all levels with various price points, features, etc.... something for everyone....thats progress!

Not sure how many hours i've slept last night (maybe a couple of months).... are we already on 3.5? sizeable improvement?

My first contribution mode nukem FSR3

link:
RTX 3080 FSR3 FrameGen A Plague Tale