Next gen lighting technologies - voxelised, traced, and everything else *spawn*

[DavidGraham]

Great news, imagine what a title designed around RT could be.

Indeed, I imagine when the hardware is more widespread, a dedicated deep RT pipeline could be created for several games, it would work like games that have dual APIs (DX11 and DX12), only this time the game will offer a DX12 path, and a DXR path.

 

[Ext3h]

You need to use more than 1 ray per pixel.
Also to my understanding it's not really set in stone 10 GRays/s but even with the RT-hardware it depends on several variables

Read it as "10 Giga-samples/s against the BVH tree". Or "10 BVH samples + >10 Giga ray-triangle intersection tests / second" if you give NVidia the benefit of doubt.

Sampling against anything but the most trivial (and best fit to the BVH) scene isn't going to achieve anything remotely better than what we see.

But that's not new, is it? Just roughly the performance numbers which had been spoiled during the "Speed of Light" talk already.


Oh, and that "low FPS with foliage is a bug" BS? Nope, if you have to check for alpha coverage, it's not for free. That was a full hit, and you have to treat it as translucent. And even on low you can not just stop there (same as with glass), but you have to treat it as translucent, or even just basic lighting / shadows break. Only way around is doing foliage with pure geometry instead of alpha.

You would notice if shadows also broke just because you cheated on objects with alpha.


Oh, and if you want something to weep, imagine DXR in a snow map. Albedo 0.8-0.9 for most parts of the scene.

So for properly working GI you don't even get off the hook easily with the 3-4 bounces "ultra" is currently (still) limited too, but even at a 10 bounces you would still be missing about 25% of light intensity. These city / forest / indoor maps, where most materials have an albedo of <0.3, are still really friendly towards GI via ray tracing. Only nasty thing they have to offer is translucent objects.

With specular reflecting surfaces, there is only one direction an incoming ray can reflect, and the reflected ray direction can be computed exactly based on the incident ray and surface normal. This works for flat, curved or bumpy reflecting surfaces. Raytracing this way results in clean reflections.

No true, unfortunately. Imagine a materiel such as brushed steel or matt plastic, viewed from a distance. The specular component follows a normal distribution.

Only polished surfaces expose the properties you are describing. When it's perfectly smooth, the normal distribution (and hence "blurry reflection") collapses into a perfect reflection.

https://twitter.com/i/web/status/1063318522112155648

Don't give too much on that either. Neither BVH nor geometry are fitting in L1 / L2. It's safe to assume that during ray tracing, the memory interface is completely limited by transaction rate. Neither latency nor bandwidth, but raw transaction rate.

 

[Voxilla]

No true, unfortunately. Imagine a materiel such as brushed steel or matt plastic, viewed from a distance. The specular component follows a normal distribution.

Only polished surfaces expose the properties you are describing. When it's perfectly smooth, the normal distribution (and hence "blurry reflection") collapses into a perfect reflection.

That's why I talked about 'specular' reflection, what you describe is 'diffuse' reflection.
In the videos I only saw specular, mirror like reflections.
BTW fake but passable diffuse reflections can be achieved by adding a bit of blur to the specular reflections.

 

[JoeJ]

With the observed 60 FPS for 1080p on a RTX2080Ti with Ultra DXR setting.
It begs the question where all those 10 GRays/s have gone ?
Assuming the best case of all pixels being reflective with a single reflection ray per pixel, this results in 120 MRays/s
This is a massive factor 83 difference between the on the box quoted 10 GRays/s and actual in game 120 MRays/s
(Or similarily paying a 500$ bonus for RTX and getting 6$ value)

We never know what scene is used for X GRays/s, so this number is useless and we can not establish it as a performance measurement.
Maybe they refer to a scene with a single triangle

I would like to know about the material complexity of BFV.
If they use hundreds of custom shaders made by artists using GUI, this could explain the bad performance maybe.
But i remember a presentation about Star Wars Battlefront where they seemingly used just a single PBR material for the scene made mostly from photogrammetry. If they still do this, shading should not be the bottleneck.
Maybe somebody with trained eyes on material shading can shed some light? (I'm too much noob here myself.)

Anyways, reflection material could be simplified against the original material if this turns out a bottleneck. Curious if we see this in the future...

Otherwise i would conclude BFV scenes are just too large. In early videos we saw the scene was heavily clipped for reflections. Entire buildings at some distance visibly had turned on and off without any smooth fade off or LOD.

 

[pharma]

You need to use more than 1 ray per pixel.
Also to my understanding it's not really set in stone 10 GRays/s but even with the RT-hardware it depends on several variables

Hopefully the BFV producer will come out with detailed information related to how they worked and developed using RTRT, eg how many rays were used in particular situations.
One thing mentioned by Nvidia at SIGGRAPH 2018 was the need for Game producers to educate the public on these new architectures.

 

[vipa899]

Dont own a RTX, not planning on getting one either for those prices as they are now. But i do like when new tech arrives, and its understandable its an early-imperfect state but that might change in a couple of years.

I wouldnt mind a RTX variant in the next gen consoles either.

 

[Voxilla]

Read it as "10 Giga-samples/s against the BVH tree". Or "10 BVH samples + >10 Giga ray-triangle intersection tests / second" if you give NVidia the benefit of doubt.

We never know what scene is used for X GRays/s, so this number is useless and we can not establish it as a performance measurement.
Maybe they refer to a scene with a single triangle

Actually by now, we do know where the 10 Grays/s come from.
Some people even tried to reproduce the number:
Basically achievable with a single BVH object with about 1M triangles, covering part of the screen, primary rays, many rays per pixel, high screen resolution (which all improve BVH cache hit in this case) and a trivial hit shader.

Not very representative for your typical raytraced game, as turns out.

 

[Kaotik]

I'm not sure wth this is, but someone's claiming BFV DXR can be enabled on Vega and it seems to work and claimed performance is ridiculous. Or are the non-RT reflections really that good in it and he's just mistaken?
https://forums.guru3d.com/threads/r...-mods-bios-tweaks.416287/page-48#post-5607107

edit:

https://twitter.com/i/web/status/1063399674239959040

 

[DavidGraham]

I'm not sure wth this is, but someone's claiming BFV DXR can be enabled on Vega and it seems to work and claimed performance is ridiculous. Or are the non-RT reflections really that good in it and he's just mistaken?

Vega doesn't support DXR or even have a driver that supports the deprecated fallback layer. He is running regular reflections. And no they don't look even close to RTX.

 

[Kaotik]

Vega doesn't support DXR or even a driver that supports the deprecated fallback layer. He is running regular reflections. And no they don't look even close to RTX.

That's what I was thinking first, but then I saw the UFD Tech tweet, they should definitely recognize the difference or go out of business

 

[DavidGraham]

That's what I was thinking first, but then I saw the UFD Tech tweet, they should definitely recognize the difference or go out of business

Yes, they are a bunch of idiots.

 

[Malo]

Yeah from what I've seen so far, the reflections are really nice and there's a clear difference between RT and regular in most circumstances. Can't really tell from his screenshots though. Since there's reflections the game is falling back to SSR so there's detection I guess, but not in the options menu?

 

[DavidGraham]

since there's reflections the game is falling back to SSR so there's detection I guess, but not in the options menu?

They are forcing the menu to show on for DXR, via ini editing .. It's a cheap trick. It's on in appearance, but not in function.

 

[Malo]

They are forcing the menu to show on for DXR, via ini editing .. It's a cheap trick. It's on in appearance, but not in function.

oh that's pathetic. Instant blacklist for stupid click bait shit.

 

[Malo]

What's the possibility or a hybrid approach to reflections? When buildings and terrain geometry is static, is it feasible to apply a high quality cube map to a window for example, then overlay Ray traced dynamic objects like animated models on top of that? I could imagine it would be hard to get look right, you would have to have the ability for rays to ignore certain pixels based on geometry and no idea if it would be a significant increase n performance to be worth it.

 

[vipa899]

They are forcing the menu to show on for DXR, via ini editing .. It's a cheap trick. It's on in appearance, but not in function.

And why would they be doing that, i mean whats their gain, just being anti nvidia, amd supporters?

 

[Ext3h]

And why would they be doing that, i mean whats their gain, just being anti nvidia, amd supporters?

Just trying if it works by chance. Application doesn't validate config / resets it to plausible, they do see reflections in game => quick conclusion that it must be working. There isn't that much of a difference between the "simple" reflections and the raytraced ones, unless you know what to look for.

 

[BRiT]

Which is akin to a bar room discussion between arm chair generals, if they can be called that. That's why is it's good to hear more informative, less emotional comments from someone on the front line with actual RTX exposure, like @sebbbi via his twitter comments. They have a tendency to put things into perspective and filter out the "fan-boy" aspect to this technological discussion.

He doesnt even have an RTX yet!

 

[BRiT]

If true, it would make the negative comments posted in this forum even worse! And especially when they are "liked" by the same "old" boys.
Do you see any negative Nvidia "jibs" in his tweets?

Every who's talking about bias and jibs instead of the topic at hand needs to be reply-banned on this thread by the mods. I don't think many would like that, because that currently looks to be all the typical PC-posters.

 

[DavidGraham]

There isn't that much of a difference between the "simple" reflections and the raytraced ones, unless you know what to look for.

SSR has obvious and clear artifacts, any object between the player's eye and the reflection surface will form a zone of zero reflections. You instantly spot this if you aim your gun at a water surface for example. A halo of zero reflections forms around the outlines of the gun.

Also SSR doesn't reflect objects off screen, so if you have a tall tower or the sky outside of your view, you won't see it reflected. You instantly spot this by looking down on a ground reflection, and the SSR reflection will not show the sky.

As for mirrors, glasses, metallic and curved surfaces, with cubemaps, reflections are not dynamic, you won't see your own reflection or your surroundings, and with SSR, reflections have missing pieces and artifacts all over. None of this happens with RTX. Finally resolution of RTX reflections is higher than regular ones.

So no, the difference is night and day even at the first glance.

At any rate DSOG tried this "cheat" and proved it presents only regular SSR reflections.