DLSS Momentum Continues: 50 Released and Upcoming DLSS 3 Games, Over 250 DLSS Games and Creative Apps Available Now

[Tinek]

TAA has been talked to death, no use splinting hairs over it.

I'm waiting for his perfect game, not just truism, a game that will actually push technology forward and be perfect from every point. Let's see how he buries DLSS and FSR.

You do realise there's no such thing as perfection? All you say here is "I don't care what it will be, I'll bash it anyway.".

 

[Tinek]

I’d argue he’s basically saying, “All cars are death traps and garbage—except for my perfect car, which I’m not going to show you in an actual game because it’s so amazing it would blow your mind.”

But that's not at all what he's saying. Also, my arguments aren't even based on what he's personally saying - I base it on modern AAA games looking like crap, not because of technologies used but because of lazy programming, bad practices, penny pinching, mismanagement etc. - whatever the case for a given title, for the monies they cost to make they should work and look WAY better. And by better I mean clear image, not the blurry mess they often implement (TAA based, but not as AA just generally abusing the tech instead of doing things properly). Again, this has been voiced by GN not that long ago and by other - games in native should never look worse than upscaled (which by definition is a lossy thing). It's like comparing MP3 to uncompressed audio and somehow the uncompressed audio is so bad quality that the MP3 of that same audio (but with AI sprinkled in) sounds better - that should never be the case and every time that is the case it's a problem. Yet, somehow, a lot of gamers don't see it as a problem, which I find silly.

 

[Tinek]

There are way way better ways to come across with a message than actually coming across as a robot and making claims of your own work yet to date not showing a single ounce of that work.

That's true, his communication skills are very lacking. In his general vids he talks way too fast, glances over details and assumes people will go and research it further themselves, but they won't. It doesn't mean he doesn't know what he's talking about, but better communication would definitely help get the point across. Any time I want to get what he's actually about, I have few pages open with additional info and run his vid in slow motion or paused to see what he's talking about - most people won't do that.

 

[Tinek]

(...)Thing is, studios don't have the resources that someone like Chris Roberts has (btw, would be nice to see a teardown of that engine ) ), so while some "quick fixes" can be had probably "easy", putting that into an actual game is something else.

Really? Are we again talking about "resources" in AAA studios that often take 5-10 years to make a game, which cost $200+ mil to make? Yet somehow "AA" and indie games with fraction of that budget and teams are able to make games that work and look just fine. Where did they money and time go, then? Or maybe someone pocketed a lot, whilst hiring inexperienced/intern devs to do bad (but cheap) programming? As often that how it looks to me. Not just visual problems, but also so many bugs and broken/unfinished parts of these games, which then require a ton of patching to fix basing things. AAA industry is simply broken badly and that's a fact. It's also not normal, yet market for some reason tolerated that for a way too long time. Gladly, things are slowly changing for the better.

 

[Tinek]

Even basic upscaling works well for 4K simply because it's an obscene amount of pixels. With something as advanced as DLSS you almost feel stupid not to upscale from even 1080p just because the visual downside vs performance benefits is so skewed in favour of the latter.

60 FPS = 16.7 ms

That data is just for CNN, Transformer is considerably more expensive computationally. Also, this in itself doesn't say anything if one doesn't also show how much time other operations take per each frame of a modern AA game. This is what TI often shows in his deep dive video showing specific games and engines - what they exactly do, how many ms it takes to computer and how it could be done differently (and how much time that would cost). I wish testers/reviewers would do such deep dive now and then - DF really should, but instead they prefer to theorise what happens in a given frame, often saying total idiocy, as they have no knowledge to actually analyse it.

 

[chrcoluk]

Recently learnt lod clamp should be set to allow globally (or at least on a per DLSS game basis), otherwise DLSS doesnt work as well.

 

[Poneros]

That data is just for CNN, Transformer is considerably more expensive computationally.

Yes, about double, check previous page for another table. Not that important anyway because people were very happy with CNN visuals too especially for 4K.

Also, this in itself doesn't say anything if one doesn't also show how much time other operations take per each frame of a modern AA game.

That's irrelevant to the discussion at hand. We were discussing if it makes sense to go to 4K given performance demands & I pointed out that upscaling makes it a very cheap proposition (and the visual quality is evident). All we need to know is native res perf (minus AA cost) vs target render res cost + DLSS cost. It's not going to be the exact performance but >90% there, so good enough to know it's worth doing.

 

[Calin Banc]

You do realise there's no such thing as perfection? All you say here is "I don't care what it will be, I'll bash it anyway.".

I'm only interested to see how his TAA implementation will best DLSS and FSR 4 (since we're in this thread).

Really? Are we again talking about "resources" in AAA studios that often take 5-10 years to make a game, which cost $200+ mil to make? Yet somehow "AA" and indie games with fraction of that budget and teams are able to make games that work and look just fine. Where did they money and time go, then? Or maybe someone pocketed a lot, whilst hiring inexperienced/intern devs to do bad (but cheap) programming? As often that how it looks to me. Not just visual problems, but also so many bugs and broken/unfinished parts of these games, which then require a ton of patching to fix basing things. AAA industry is simply broken badly and that's a fact. It's also not normal, yet market for some reason tolerated that for a way too long time. Gladly, things are slowly changing for the better.

That's not just performance, but also gameplay and just features in general. Part of the cost would be the high salaries in USA and Western Europe.

Some improvements aren't cheap to make. I've mentioned Chris and Star Citizen since they needed 2 years + to proper thread and change the engine to Vulkan and that with people that worked at Crytek. They kept on improving and optimizing certain aspects, but it ain't always easy and cheap.