Well they were too far in development when DLSS was introduced and it was by the rival vendor to the one they contracted to.
I don't think it's a mistake, in that context.
-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
-
We have made minor adjustments to how the search bar works on ResetEra. You can read about the changes here.
Was it a mistake for new consoles to not have DLSS like technology?
- Thread starter stoke1863
- Start date
I don't think it's a mistake, in that context.
There are various reconstruction methods already in use and being improved upon, AMD will also have their own solution soon.
The restriction is part of the reason I think it is probably non-trivial to implement DLSS and get good results. The reputation DLSS has right now is based on Nvidia tightly controlling who gets to use it, and probably also providing direct engineering support.
I suspect that if Nvidia started allowing anyone to use DLSS, that we would get a lot of crappy implementations with bad image quality and artifacting, and people's general perception of DLSS would get more negative. At least, that's the only explanation I can come up with for restricting access.
You're right of course, but I think there's a good chance it's because it's still in development rather than because it's fundamentally flawed and will always require Nvidia supervision to implement.
Remember that 'DLSS 2.0' hasn't even been around for a year yet (and it was a pretty rough year at that).
DLSS seems very impressive for the games that make use of it.
I'm curious as to what will happen when AMD comes up with their own version. That coupled with Smart Access Memory could prove to be even more impressive.
I'm curious as to what will happen when AMD comes up with their own version. That coupled with Smart Access Memory could prove to be even more impressive.
They will only be ahead if devs gimp those multiplatforms. Most haven't in general I just drag and drop redoing symlinks or take big files the game consistently needs. It's not effecting current multi platforms released this year.
Even then it means little if you're using a ram drive like myself which is still faster than PS5 solution at 3800 at cl16 in a 64GB solution. Windows rarely wants more than 20GB so the rest is for the games, production, or networking.
PC gamers aren't restricted to SSD. This advantage will be wiped without a doubt when DD5 shows up based on secs and having DD4 in front of me.
We are likely to see DLSS and nvidia's sam solution before AMD has a viable solution to DLSS combined with their sam tech.
Maybe, maybe not.
I just hope AMD's answer to DLSS allows for use with any game, and not specific ones. Even if it weren't as good, I'd rather the ability to use it with any game I want.
They don't need to train on every game with dlss 2.0 though right?
I can vibe with that either vendor having a great upscaling solution is great to expanding their markets or making aspects of high end gaming more viable for people without tons of cash to spend.
AI based reconstruction is definitely the future (and present), but a console can never possess every cutting edge tech when it launches.
I'm damn impressed and very grateful that the console RT solution is as good as it is. I was worried consoles would be going completely without RT all this gen, which would slow PC adoption as well.
I'm damn impressed and very grateful that the console RT solution is as good as it is. I was worried consoles would be going completely without RT all this gen, which would slow PC adoption as well.
From what I read about DLSS, there is nothing special on the pc/console side, all the magic(training) is happening on the NVidia's server. I think it's possible for MS/Sony to add client side code for DLSS-like solution, if they and amd want to invest in the server side of the equation.
The point is that it's coming and it's hardware agnostic. DLSS 1.0 was 2/10 as well.
CBR reconstruction was invented by Ubisoft, TAA supersampling - which is basically what TAA injection/reconstruction is - was invented by Crytek I think? Both happened years before the release of PS4 Pro.
The thing with ML based reconstruction is that there's very little point in each studio maintaining their own NN for such reconstruction as it would be mostly universal. Thus a more possible scenario would be a middleware solution which will be made and maintained by either the platform holder or the engine provider (Epic, Crytek, DICE, etc)
They don't but further training and NN improvements will still lead to better results. This is partially why they've added the Ultra Performance mode to DLSS 2.1 a year later.
So hardware agnostic that they've used an RTX GPU with tensor cores for that presentation.
That presentation was nearly two years ago.
Console generations are usually 6-7 years. There's time to 'come'. But you already knew that.
Came into to give a concise answer, and saw you already did such
absolutely. three years from now it will be a BIG mistake and it will only get bigger.
obviously it wasn't necessarily available since the consoles went with AMD, but it's a dark shadow that will hang over this entire generation.
obviously it wasn't necessarily available since the consoles went with AMD, but it's a dark shadow that will hang over this entire generation.
There is this assumption that DLSS is 100% on Switch but is there anything that firmly confirms that even at all or a hint? DLSS is not free like many said. It involves adding hardware cores to the chip thus increasing it's cost to manufacture and for what? Some games that may utilize it or need it? We're talking Nintendo here and I don't think tensor cores that is only beneficial to just certain games is expected of them.
And stop the DirectML stuff. It's an API utilizing whatever it gets it's hands on for the sake of DirectX on PC.
And stop the DirectML stuff. It's an API utilizing whatever it gets it's hands on for the sake of DirectX on PC.
I should have said "most AAA games". It's also going to be a feature of Unreal engines afaik so a lot more games could support it.
Where did you get that from ?
According to this reddit post it's much less but I don't know how accurate it is
RTX adds ~1.95mm2 per TPC (tensors 1.25, RT 0.7) : hardware (reddit.com)
Checkerboard rendering is still a great solution IMO, all previous comparison between checkerboard rendering are 30fps, if the dev target 60fps, there would be twice as many frames blending in a second and less image breaking and result of better image quality in motion and less artifacts. If you look at ghost of tsushima, all the ghosting artifact from moving grass from TAA are significantly reduced. Other imagine breaking artifacts from checkerboard rendering should also be reduced.
Similar estimation.
You have to remember that it's more than just the processing h/w itself which is needed in a GPU to add new units though.
You have to add new data paths which leads to adding more on-die storage and beefier caches.
The processing h/w itself is in range of 20% for RT+TC with TC being the bigger consumer here but if you account for the changes required to actually add this h/w into the GPU then it will be more than that.
So ~20% for TCs alone seems like a close enough estimate.
Then again this is based on Turing and we don't know how this has changed with new RTCs and TCs in Ampere.
Yeah and also they can render at a higher base resolution which also helps to make it come out clean.
No. It would have to be a total different architecture if they were replicating what nvidia is doing. People will need to wait and see what AMDs solution will be.
Yup, as long as the pixel match the screen like 2160p checkerboard instead of 1800p checkerboard. I think days gone looks cleaner than ghost easily and I believe its still using a 2160p checkerboard setup.
What I am talking about are games that have core game design elements that are dependent on being able to load data from the SSD very quickly such as the PS5's upcoming Ratchet & Clank game, not quicker loading screens.
Lmao don't you have to be some ai expert to do dlss? Microsoft is working on a solution for the xbox, seems way off though.
I'm not talking about the company that invented it (though if you keep dropping the quality bar a basic bilinear upscale coudl be seen as reconstruction), I mean Sony was really the first to start using it at such high quality levels as to look almost native. Spiderman or God of War for example look very close to a native 4k on a PS4 Pro. It's not as good as DLSS which does similar feats at quarter resolution instead of half, but it's still impressive.
Any gamer get the benefits of solid IO from DD4 in any decent streaming engine besides loading screens. You're OS runs better too. PS5 games aren't the first to benefit from decent IO streaming engines are.
I'm welcome more dev flex in functionality but loading screens aren't the only benefit to solid IO peformance when you got it.
This is the first gen where almost no platform will be weak in this area for once.
Isn't AMD working on a DLSS competitor that will support RDNA2? If that's the case, then there's no reason it shouldn't support the Xbox consoles at the very least, and should most likely work on the PS5