Impact of nVidia Turing RayTracing enhanced GPUs on next-gen consoles *spawn

Status
Not open for further replies.
At quarter resolution with no AO, lighting is toned down, and no reflections within reflections.
Are we realistically going to have those qualities in Realtime raytraced reflections in potential console hardware? 2 samples per pixel is very noisy. Denoising fills in the geometry but can't resolve the texture detail. Hence you'll be getting the equivalent of reduced resolution and possibly no AO, depending on how many recursions can be handled.

The important point in this comparison is that rasterisation hacks should not be compared to the best raytracing has to offer, but to whatever raytracing can achieve within the limits of its implementation in a console (an unknown at this point). Hybrid renderers will be required, and if you can't combine the two comfortably (curved surface reflections? Can you rasterise the texture detail and trace the light?) you'll have to have compromised tracing.

It's just occurred to me that multi-pass tracing may be more effective, the same as the deferred rasterising. You could trace geometry and bumps in full resolution, one sample per pixel, and get a complete, noise free (though aliased) geometry, and then trace noisy light to combine with that.
 
Last edited:
At quarter resolution with no AO, lighting is toned down, and no reflections within reflections. Also curved surfaced lack dynamic reflections altogether, and rely on cubemaps.
yes, it is always a compromise, but it is already possible on low end hardware. This is what I meant when nvidia showed their demo with the flame-thrower reflections and did not use any rasterization trick to get at least a decent reflection without RTX. It is possible. It needs some more power, but it is not impossible.
btw, it may be at quater res but the xbox x renders in 4k and the reflection is still 1080p. That is the res the demos were running on multiple nvidia cards. You can have much better reflection quality with that much power (one of those cards has more than twice computational power). It may not be as good as ray-tracing (with all it's special hardware) but still be convincing enough.
 
Is it possible to "artificially" gimp or cull near future performance of gpus with AMD/Nvidia Drivers to have better segmentations?
 
At quarter resolution with no AO, lighting is toned down, and no reflections within reflections. Also curved surfaced lack dynamic reflections altogether, and rely on cubemaps.
You should check DF's analysis of RT battlefield V again to see the render quality of reflected objects. What Hitman is delivering here is arguably more.
 
If all you need is one good planar reflection, rasterization will probably be a performance win, even on RT accelerated hardware. Maybe even 2 or 3 different planes (with well optimized reflection culling and etc) will still be more efficient brute-forced with extra cameras, as hitman is doing. RT becomes a win when you have "infinitely many" arbitrary reflection directions.
Still, hardware RT accelerates the Ray traversal and intersection test. Once it hits a surface, it has still got to be shaded, and shading it's not free. You reflection samples will be competing with your "primary rays" samples (the traditional rasterized main camera view of the scene) for shading time. So don't expect full quality/full res there. Even recursive inter reflections aren't very likely, at least not without a sensible limit. That demo by that Soviet game was misleading in how it trivialized those costs.
 
Last edited:
Planar reflections on consumer hardware are as old a dinosaurs...

22 years ago: Valley of Ra demo on 3dfx Voodoo 1:
BE1F984.jpg


Source Engine since Half Life 2 (water).

Unreal 1 20 years ago..

Rockstar's RAGE engine since Max Payne 3 which was full of them (and since then GTAV & RDR2)..

Every single engine under the sun support them...they are just not used much because of perf hit (now is it as big of a hit as RT...this would be interesting got see) & only work on planes...
https://docs.unrealengine.com/en-us/Engine/Rendering/LightingAndShadows/PlanarReflections
 
Planar reflections on consumer hardware are as old a dinosaurs...

22 years ago: Valley of Ra demo on 3dfx Voodoo 1:
BE1F984.jpg


Source Engine since Half Life 2 (water).

Unreal 1 20 years ago..

Rockstar's RAGE engine since Max Payne 3 which was full of them (and since then GTAV & RDR2)..

Every single engine under the sun support them...they are just not used much because of perf hit (now is it as big of a hit as RT...this would be interesting got see) & only work on planes...
https://docs.unrealengine.com/en-us/Engine/Rendering/LightingAndShadows/PlanarReflections
I'm not sure that 3Dfx demo uses planer reflections. Unless I'm mis-understanding how they work, planer reflections basically render the reflection in realtime by capturing a scene from the perspective of the reflected surface and render that out to a texture. That 3Dfx demo is clearly rendering geometry for the reflection, and if you look at the wireframe, the reflected models are lower polygon than the actual ones.
 
I'm not sure that 3Dfx demo uses planer reflections. Unless I'm mis-understanding how they work, planer reflections basically render the reflection in realtime by capturing a scene from the perspective of the reflected surface and render that out to a texture. That 3Dfx demo is clearly rendering geometry for the reflection, and if you look at the wireframe, the reflected models are lower polygon than the actual ones.
You can render to texture or render reflection within the viewport. (Using stencil.)
Having proper LoD and reduced shading is a good idea. (as is some form of antialiasing especially if rendering in low resolution.)
 
At quarter resolution with no AO, lighting is toned down, and no reflections within reflections. Also curved surfaced lack dynamic reflections altogether, and rely on cubemaps.
If we don’t recognize the merits of rasterization then people are unlikely to recognize the merits of ray tracing.

I caution stepping into bounds where just because a superior solution exists it immediately makes anything else completely unacceptable. In this scenario, the planar reflection is a win for this particular setup. A setup where scene complexity is minimal and contained, and has very little really happening.

If all you need is one good planar reflection, rasterization will probably be a performance win,
These don’t work everywhere for every game, if the only time we get to see good reflections is to setup the scene for that particular feature then games would suck. It’s why we see much more liberal use of SSR than planar.

It’s nice to see an example of a rasterization win, but the limitations on where and how it can be used is what puts RT ahead.

I very much disagree with the concept that because title X has this feature it could be integrated to every single game and engine out there.

And this is where the stance of RT should be. Granted today in 2018 it’s true we are likely to only see it be used in the high end. But years from now these features will trickle down to the mainstream players and consoles should be,imo, part of that group if they are launching 2020+

We have yet to see a title built ground up with RT in mind from the bottom up.
 
You can render to texture or render reflection within the viewport. (Using stencil.)
Having proper LoD and reduced shading is a good idea. (as is some form of antialiasing especially if rendering in low resolution.)
Would it still be represented as polygons in wireframe mode, though? Also, I don't think the floor reflects anything other than the characters. It's been a while since I've seen that demo running in real time, it came with my Orchid Voodoo 1 back in the day, but I'm pretty sure it's just rendering models under the floor and not reflecting anything else.
 
Yeah. RT reflections are in a world of their own. Planar that works on the floor won't work correctly for shiny table-tops that are at a different height and so different perspective, and completely fail for a shiny vase on that tabletop.
yes, this is true, but question is, if we need to use all that power the 2080 offers just to get better reflections, if even a ps4 pro can do it properly in most cases.
And btw, most people won't see any difference, because you don't stand in that game and watch, you play those games. So nearly perfect reflections in a vase are one thing, but are they really necessary. It seems to be the first thing I would optimize in order to get better performance.
Less reflective surfaces means better performance, because if something reflects, the rasterizer musst still calcuate those things.
Reflections in a window like in the nvidia BFV flamethrower demo are possible with just the trick hitman 2 uses here. An really old trick but it works with much less power (yes I know and less precision).

the problem with extra hardware still remains, if they build something on the chip that needs extra-space, there is almost only one thing that can be done with it. A compromise and more flexible solution that can also be used for other things is much more favorable. And btw, for current RT-hybrids you still need much more rasterization power. Both would be really bad for console hardware if it isn't flexible.
 
The flame in the window can be done with planar reflections. It gets more complicated with the car - you'd need a cube-map. The most significant problem though is developing solutions for each reflection, whereas raytracing provides a unified solution that'll work for all objects. the simplification of the developers lives in that respect is probably worth any drop in reflection quality.

There's a lot of discussion already about more flexible solutions, and AMD's solution seems to go more that way. There's also a problem in RT's portrayal at the moment as only a graphical advance. The inclusion of ray-tracing accelerating structures could well benefit other areas of games like AI. The bigger picture of custom console hardware could be a lot more balanced than RTX is appearing, with useful, versatile acceleration structures.
 
yes, this is true, but question is, if we need to use all that power the 2080 offers just to get better reflections, if even a ps4 pro can do it properly in most cases.
And btw, most people won't see any difference, because you don't stand in that game and watch, you play those games. So nearly perfect reflections in a vase are one thing, but are they really necessary. It seems to be the first thing I would optimize in order to get better performance.
Less reflective surfaces means better performance, because if something reflects, the rasterizer musst still calcuate those things.
Reflections in a window like in the nvidia BFV flamethrower demo are possible with just the trick hitman 2 uses here. An really old trick but it works with much less power (yes I know and less precision).

the problem with extra hardware still remains, if they build something on the chip that needs extra-space, there is almost only one thing that can be done with it. A compromise and more flexible solution that can also be used for other things is much more favorable. And btw, for current RT-hybrids you still need much more rasterization power. Both would be really bad for console hardware if it isn't flexible.
Because not all engines are equal. And not all engines can use the latest and greatest methods and work around methods. But all engines could adapt to including DXR. Some games, like Bethseda titles, their engine is dedicated for other goals, but they don't want to sit in the stone age forever with their graphics. Looking at Fallout 4 and 76, and now Starfield. They all use Creation, and Creation is hurting in all sorts of areas, including lighting, shadows, AO and reflections. Even though Ubisoft has snow drop, they still have to support Dunia, and Anvil. All of them doing different things. Content creation is incredibly more important than graphics, and there are times in which the way they build worlds and levels will conflict with rendering. That's why I don't like this concept of applying AAA 1P adventure titles, and claiming them as graphical king and using them as a baseline for all other titles, when they are designing the entire game around the limitations of these features.
 
These don’t work everywhere for every game, if the only time we get to see good reflections is to setup the scene for that particular feature then games would suck. It’s why we see much more liberal use of SSR than planar.

I thought I had made it obvious that I know that.
I just wanted to point out that in the hypothetical scenario that all you need is one plane of reflection at all, rasterizing a second camera will be faster even on current RT acceleration solutions. If that were not the case, they would be forgoing rasterizatuon and hybrid rendering all together right now and just use raytracing for primary rays as well as everything else.
From that conclusion, I dare say the same is probably true even if you have a couple discrete reflective planes or more. Obviously that only takes you so far. If using RT, what Hitman could have done was have all reflective surfaces reflect the entire scene all the time, which is obviously better, given the resources to achieve it. Untill then, I commend their brave choice to try to pull that off now. There is something very dramatic about seeing your own reflection on the window of a house you are about to invade. It was the right trade-off to spend all the perf.bthey must be to do that there.
 
Would it still be represented as polygons in wireframe mode, though? Also, I don't think the floor reflects anything other than the characters. It's been a while since I've seen that demo running in real time, it came with my Orchid Voodoo 1 back in the day, but I'm pretty sure it's just rendering models under the floor and not reflecting anything else.
If stencil method is used, yes the wireframe should be visible.

Agreed that they most likely did the transparent floor and flipped characters trick.
 
Status
Not open for further replies.
Back
Top