Carmack on Ray Tracing & Rasterization

I disagree somewhat, what you described is at best a hybrid title, a next gen featured game with current gen assets or current gen asset game with next gen features. For me a pure next gen title comprise of the vast majority of its features and assets that can not be run on a current gen console. I mean you wouldn't call all those Nvidia Gamework featured multiplatforms next gen titles would you?
With the way consoles have moved into the PC space. This argument doesn't work as well anymore. We can scale games down to oblivion if we need to.

Ray tracing will never be considered 'low' graphics tier anytime soon.

Keep an eye out for Halo Infinite and Forza 8 - I think these will be next generation, but will be scalable to this generation.
 
What if next-gen assets are no different to this gen? We're already at a very high quality. What if next gen is nothing more than this gen but with with photorealistic lighting? Would you deny it was a new generation at all?
Well that would depends on how photorealistic we're talking about :), we're not getting nearly enough number of rays in games like Metro or Tomb Raider which subsequently look either too dark or worse than the rasterizer sometimes. On another hand the asset quality could still be vastly improved, a big focus on photogramatry by getting much higher fidelity scanned assets into game engine even with traditional lighting is a huge win. The end result is decisively better overall.
Metro RTX Ultra
https://www.purepc.pl/image/news/20...zamy_dlss_po_nowej_aktualizacji_gry_nc1_b.jpg
Book of the Dead
https://blog.megascans.se/wp-content/uploads/2018/01/3.jpg
One still looks gamey even with raytracing on while the other is very CG esque.
When ray tracing isn't just a bolt on to get people to purchase your software, then, yes I would start to consider us in the next gen. Ie; when the next consoles are deployed and we see the features in full constant use. This is a transition period at this moment (but still mainly marketing), but I think as we see with Control there will be a variety of ways to implement ray tracing into our existing stack, the launch methods employed (bolt-ons) are likely the easiest and fastest way to get ray tracing into the game, but at a huge penalty to performance. It's nice to prove that brute force works, but it's about as elegant as Super Sampled AA. We know that's the best quality, but no one uses it.

As developers spend more time to use ray tracing in conjunction with our current T&L techniques (perhaps just to augment them to handle edge cases in which traditional methods are too costly to employ), we will get high performance hybrids which I suspect is where we all want to be for next gen.

Ray tracing is an effective tool for developers to hybridize what they are working on, the idea of full screen real time ray tracing was never the intention and people hoping that would be the norm will be disappointed.
I'll just have to wait and see what a ground up RT designed game would look like and at what cost. I do hope for the best tho, the last thing I want is for such a feature to be implemented but severely limiting advancement in asset, FX, simulation etc.
 
To get much better geometry quality hard problems will need to start being tackled sooner or later, we need continuous geometry LOD at some point. Would be nice for foveated rendering too.

What hybrid VI and raytracing shows is that random sampling and smart anti-aliasing just can't solve everything efficiently ... it's relatively easy and as such attractive, but no matter how hacky and ugly things get, pre-filtering remains necessary for bests results.
 
Well that would depends on how photorealistic we're talking about :).
It's a rhetorical test of your definition. You say any game using current-gen assets is current gen. Hypothetically, if next gen is this gen assets but at truly photorealistic/cinema CGI level rendering, impossible on current hardware, by your definition they'd still be current-gen games. Ergo, do you stick with that definition or look for another one thus requiring you to re-evaluate your position on what constitutes a next-gen game?
 
It's a rhetorical test of your definition. You say any game using current-gen assets is current gen. Hypothetically, if next gen is this gen assets but at truly photorealistic/cinema CGI level rendering, impossible on current hardware, by your definition they'd still be current-gen games. Ergo, do you stick with that definition or look for another one thus requiring you to re-evaluate your position on what constitutes a next-gen game?

Is this even an important distinction to be concerned about in the first place?
 
Only in so far as talking about next gen games and features is, or isn't, important. ;) If we don't care whether next-gen is a next-gen or not, it doesn't matter. If we want to talk about the difference between next-gen content and current-gen content, we need to know what makes something next-gen. If the qualities of 'next-gen' are present in some current titles on some current hardware, that'll be a next-gen game rather than a current-gen one, and the comparisons to future console workloads may not be expected to be any different.
 
It's a rhetorical test of your definition. You say any game using current-gen assets is current gen. Hypothetically, if next gen is this gen assets but at truly photorealistic/cinema CGI level rendering, impossible on current hardware, by your definition they'd still be current-gen games. Ergo, do you stick with that definition or look for another one thus requiring you to re-evaluate your position on what constitutes a next-gen game?
Put it simply, the Raytracing quality for games is not remotely close to what Pixar is doing with their RT and it's unrealistic to expect next gen consoles to reach that quality. The entry level RT we'll come to expect for next gen consoles alone does not warrant a next gen claim due to not enough wow factor compared to Pixar level. I guess what constitute to a true next gen game should be how drastically better the end result looks compared to current gen games, it's the sum of its parts.
 
It's not unrealistic to think that next-gen lighting using new, hybrid approaches won't be significantly advanced and different to what this generation can pull off though, and to do that, if the asset quality doesn't improve (instead just having far more variety perhaps, and quality in terms of fabric simulation and things unrelated to data resolution), that won't change next-gen being a clear generation forwards. Where you say current RTX cards are struggling to achieve good quality results on 'current gen games', the techniques being pioneered on them, which presently aren't super efficient but which have shown considerable improvement, aren't at all current-gen, and I don't think one should assume that next-gen is going to throw even higher quality assets than this gen such that mediocre performance on RTX cards == poor performance on next-gen games.
 
It's not unrealistic to think that next-gen lighting using new, hybrid approaches won't be significantly advanced and different to what this generation can pull off though, and to do that, if the asset quality doesn't improve (instead just having far more variety perhaps, and quality in terms of fabric simulation and things unrelated to data resolution), that won't change next-gen being a clear generation forwards. Where you say current RTX cards are struggling to achieve good quality results on 'current gen games', the techniques being pioneered on them, which presently aren't super efficient but which have shown considerable improvement, aren't at all current-gen, and I don't think one should assume that next-gen is going to throw even higher quality assets than this gen such that mediocre performance on RTX cards == poor performance on next-gen games.
Hey you never mentioned a hybrid solution till now you shifty bastard :). Joking. But yeah that might be what we need to balance things out, RT brute forcing might be just too much to handle even with proper optimization I reckon. I'm welcoming any surprised tho, maybe some talented devs could pull off something out of the blue, we'll just have to wait and see.
 
I was talking about what was seen on screen and not how that was achieved. I never said the result had to be perfectly raytraced, but that it looked like a Pixar movie. That was the point of the rhetorical question. ;) Just because current rendering on RTX isn't particular next-gen, doens't mean the potential isn't there and won't be realised. I think we all agree that if next-gen lighting could be 'next gen', we'd be happy with that as a new next generation, regardless of whether we hope or doubt or expect that'll pan out. In some cases, like yours, you are so doubtful of the lighting improvement ever happening that you want everything to focus on what we already have in terms of asset quality and content and don't want that to hit a roadblock while the rest of the rendering only gets a marginal improvement.
 
so... John Carmack says he wouldn't mind working on some of his classic IPs.

Last week at an Oculus event Carmack said he had done some work on an existing open source Doom 3 conversion for Oculus Quest. Before sharing it he sent a formal request to Zenimax to give him explicit permission, but they blew him off.

While I’m sure the scope of this would bleed into other things, I think that’s the context for this tweet.
 
Back
Top