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

[Deleted member 2197]

Better, in fact, as you won't have laggy lighting updates with the light from a plasma beam illuminating your cockpit three frames after the thing hit.

So is it now confirmed that laggy lighting updates expected in RT scenarios or is it still conjecture? Makes you wonder why studios are moving forward with RT games.

 

[jlippo]

Pretty pointless application of RT though. Very little bounce lighting. Space is actually one of the genres where fakes can be exceptional and as good as real RT. Better, in fact, as you won't have laggy lighting updates with the light from a plasma beam illuminating your cockpit three frames after the thing hit.

Replacing shadowmaps with RT must have been easiest decision ever.
Sharp and very long visibility range is not good combination for them.

Faking GI interreflection for for space ships is something I have been wondering many times. (Perhaps some form of classic PRT or probes around surface..?)
I'm not sure if anyone had tried it, even though interreflections within ship would be the strongest source of bounce light in dark of space.
Most games just seem to use cubemaps to get some additional light.

RT should make it quite bit easier as well, there should be amazing amount of shortcuts one could make. (From very simple shading/objects for intetreflection to distant objects being silly proxy objects with average lighting in vertex colours.)

 

[Shifty Geezer]

So is it now confirmed that laggy lighting updates expected in RT scenarios or is it still conjecture?

It seems to be the current state-of-the-art. We need a way to either get more samples, or denoise, and denoising presently requires taking samples over time, which means laggy updates. It may be a cleverer approach to lighting appears to solve this. I'd be happy to see 'low-res' immediate updates resolving into superior clarity over time if necessary, I think. Things like a fireball don't need super-sharp lighting or perfect secondary bounces to look the part, but they do need the lighting to keep up with the ball's position.

 

[JoeJ]

Anyway, https://devblogs.nvidia.com/rtx-best-practices/ has a lot of indirect information about how RTX works under the hood.

Assuming reordering is only worth it for incoherent rays, there likely would be a flag for the pipeline to turn it on or off by the user.
But there is no such flag, and the current focus may be on coherent shadow / sharp reflection rays, and there is ImgTecs claim, so... ok, i assume there is no reordering yet.

If laggy light is part of real-time raytracing, I'm starting to take umbrage with calling it 'real-time'.

Lag is necessary because processing power is finite. I don't think we can ever get rid of it completely, and also it would be new to me NN based approaches would need no form of history.
For GI the information from a single frame can't be enough to produce a realistic approximation.
It's not the noise that causes the lag, it is the infinite amount of calculations we would need to solve all those reflections. (My approach causes no noise, but i have lag too.)

But i don't think it's a big problem, personally i have those features to limit visual lag:
Cache directional lighting in worldspace, so no ghosting, and view dependent specular keeps angular correct even under motion but outdated lighting information.
More updates at regions where more changes happen.
Temporally reducing detail if overall change is high, so more surface area can be updated in the same time (e.g. if all lights turn off).

This helps a lot already, but i expect much more from calculating direct lighting with lag free traditional methods, which is a feature i don't have yet.
Basically only indirect lighting (or area lights where traditional stuff sucks) would cause lag, and if that's visible at all it's just the price for realistic lighting.

So with some engineering effort this problem is surely solvable good enough. It also helps that real world moving lights usually can use traditional point / spot lights (cars, flashlight).
I'm more worried about specular artifacts and limitations than about lag.

 

[Shifty Geezer]

But i don't think it's a big problem, personally i have those features to limit visual lag:

I expect there to be quite some developments in this area. I get the sense that there are actually a lots of fields and approaches that can be used, with various cached and approximate solutions, potentially using raytracing to shift a lighting solution towards correctness rather than to build it from the ground up.

So with some engineering effort this problem is surely solvable good enough.

The little I've seen, I can't say I agree with you. I think there's an uncanny valley response happening. As the lighting approaches realistic, the deviances become more pronounced. When you can see a video of the real world, but the lighting is trailing what's happening by half a second, it'll look 'freaky' on an emotional level. "This is supposed to be real, but it's not real. Or is it? Wrong wrong wrong warning alerts! Keep an eye out for the weirdness and work out what's happening!!"

Unlike other artefacts like weak shadows, pixelation, pop in, etc. that are just nuisances we switch off from, I can't stop my brain from noticing and focussing on delayed shadowing and such. I think it's nearly as pronounced for me as audio latency.

 

[JoeJ]

I think there's an uncanny valley response happening.

I think the same. The thought came up after watching this:

Somehow that's very realistic models and textures, lighting is likely pretty correct too (too dark because low number of bounces maybe).
But it looks uncanny to me. Almost real, but wrong. I think it's the material shading - likely i was naive when thinking standard UE4 PBR would be good enough for most stuff.
So even this is a different problem than lag, i'm worried now everything becomes uncanny, not just characters.

To avoid this, likely we need to keep art direction with games in mind, not yet realism. Recently i saw that new Rats game, and Rage 2 right after that. Rats aims for realism with all those quixel textures, but i was not impressed from the visuals. Then Rage 2 was super impressive - it looked consistent and just great to me.

I can't stop my brain from noticing and focussing on delayed shadowing and such. I think it's nearly as pronounced for me as audio latency.

It's interesting you are so sensitive to that. I agree half a second is too much, but could you list some examples where you noticed this?
Or is it just about the UE4 path tracing which surely is unpractical for games, while something like Quake2 RTX is ok?

 

[OCASM]

The NVIDIA denoiser has a "spatial-only" mode. It just doesn't look as good as the spatio-temporal one.

 

[OCASM]

https://forums.unrealengine.com/com...uter-space-playable-dxr-ray-tracing-tech-demo

 

[iroboto]

We will have to keep an eye on what they did with Call of Duty; This was a Geforce RTX ad that showed up on my FB feed. The trailer below is running realtime on a 4Pro, so yea, you wont' spot anything DXR in there.
***
The rules have changed.
Excited to see DirectX Raytracing coming to Call of Duty: #ModernWarfare

 

[cheapchips]

Df have an article up.

https://www.eurogamer.net/articles/...are-delivers-series-most-advanced-visuals-yet

 

[Malo]

Because "in-game" trailers have always been representative of the actual released game, especially by AAA publishers...

 

[iroboto]

Because "in-game" trailers have always been representative of the actual released game, especially by AAA publishers...

It's funny you wrote this, because I just finished reading the DF article. lol

From a technological perspective, the trailer demonstrates a radical revamp of COD's rendering technology and the trailer revealed today is running in real-time on PlayStation 4 Pro. No CG fakery, no 'in-engine' footage - what you see is what you get, and it's a new milestone for the franchise.

 

[Shifty Geezer]

There's still a tangible difference between cut-scene and gameplay though. eg. You can prebake everything in a realtime, in-engine cut-scene because you know exactly where it'll be. The moment you enter fully dynamic gameplay and can't use those techniques, you have to use realtime techniques which are never as good. Or basically, we should judge the new engine when we have user-created gameplay rather than commercial cherry-picked.

 

[Malo]

It's funny you wrote this, because I just finished reading the DF article. lol

None of it was the actual game though, just transition scenes? Yeah, it looks like a very nice upgrade in lighting and rendering for their engine. But haven't we been here hundreds of times in the past?

edit: what shifty said.

 

[iroboto]

None of it was the actual game though, just transition scenes? Yeah, it looks like a very nice upgrade in lighting and rendering for their engine. But haven't we been here hundreds of times in the past?

edit: what shifty said.

i don't disagree, it's the commentary from Richard Leadbetter that made me ask if perhaps he was referring to this being real-time in game. Specifically the term no 'in-engine' footage. I would normally describe a lot of games doing 'in-engine' footage, like battlefield etc. But this is the first time I've seen Richard specifically call out that it's not. He ends with, "what you see is what you get"

 

[OCASM]

We will have to keep an eye on what they did with Call of Duty; This was a Geforce RTX ad that showed up on my FB feed. The trailer below is running realtime on a 4Pro, so yea, you wont' spot anything DXR in there.
***
The rules have changed.
Excited to see DirectX Raytracing coming to Call of Duty: #ModernWarfare

I thought the night vision scenes were real life footage for a moment there

 

[Malo]

i don't disagree, it's the commentary from Richard Leadbetter that made me ask if perhaps he was referring to this being real-time in game. Specifically the term no 'in-engine' footage. I would normally describe a lot of games doing 'in-engine' footage, like battlefield etc. But this is the first time I've seen Richard specifically call out that it's not. He ends with, "what you see is what you get"

If it's not FMV (which it's obviously not), there's really no difference between "in-game" and "in-engine", is there? Cutscenes are going to run at a lower fps and have different effects applied.

Is it impressive? Hell yes, the realism is amazing.

Is the trailer and the DF article giving the impression that gameplay would look like that? Yes, they do. And that's likely not the case. That's all I'm saying.

 

[AlNom]

The night vision bits looked great.

The helicopter fly-around at 1:10 looked very game-y

 

[milk]

there's really no difference between "in-game" and "in-engine", is there?

Technically there is. Developers have historically used the specific term "in-engine" when the footage shown was rendered with their current (normally WIP) renderer bit it's not really hitting the framerate at which the end footage is displayed. They render the frames at the pace they can, and make a video of it aftewards. Most of the time, it's not ill intended, game dev does take time, and optimisations keep happening throughout the cycle. They usually use targets they think they are likely to hit, but even then, parts of the optimizations they have to do include simplifying graphics and cutting features.

 

[milk]

Df have an article up.

https://www.eurogamer.net/articles/...are-delivers-series-most-advanced-visuals-yet

Having read it: meh.
Feels like an article handed by Actvision/Infinityward pretty much written already, and Richard may have changed a few words here and there. It's more of a PR plug than a typical DF investigational piece.
Surely, it's ok for DF to collaborate with devs to get insider's info. In fact, that is absolutely desirable. But I feel like the info presented here was not put into the context you'd expect from something coming from Richard's mouth himself. Most features described are prety much standard fare for modern AAA engines. In fact, many of them were already bullet points of previous CoD. They are probably being improved uppon, but then they should be presented as such, instead of being implied as some new revolutionary addition. Even for 60fps games, most features are not completely new there.
Again, DF would not know most of the info there if it were not given to them by the dev's themselves. You can't deduce the technical underpinnings of an engine from a short trailer like that, so it's cool that DF landed the confidence from the devs to get that sort of info from them, but it's just lame that they passed it forward more as an advertisement than an actual informative journalistic article, with proper contextualization and less hyperbole.
I also don't want to diminish the devs work. Even if they are improving on previously laid out work, or incorporating things other games have done before, that does not cease to be an improvement for their game, I just wish the article had presented it more as such.