Experimental support of parallax mapping in 2.13

[HMart]

6 hours ago, stgatilov said:

I'm not sure this is true.
Even with per-pixel raytracing, POM seems to be faster than throwing more polygons.

Perhaps you want to test it yourself?
It would really interesting to see competition between parallax maps and high-poly + normals maps (perhaps tessellation shaders also?)

But it is indeed easier to just throw more polygons at the problem, no need to dance around the quality issues.
Today's "cutting edge" approach is to use shitload of polygons with auto-generated LODs and hope that GPU does not die rendering it (UE5 Nanite).

"POM seems to be faster than throwing more polygons"  Good to know but I did heard the contrary from some people in the gaming industry but I never tested it my self and this was years ago, so things most have changed by now. 

But even so even if POM is faster and look awesome when it works, there's no denying that it has serious issues with curve surfaces and characters.

And your totally right about UE5 nanite, IMO this focus on that tech, is creating huge performance issues for modern games, some developers are even stopping making custom LOD's anymore and relying entirely on nanite.

And is not only nanite, mesh shaders are gaining popularity and they gave rise to D3D12 "meshlets", that is what Alan Wake 2 uses for example.  The Remedy engine debug visualization of them, does look very similar to that of nanite, so perhaps they are the same thing? 

Lumen is another thing that IMO is creating big performance issues and ugly noisy games, even TAA cannot solve that, I played a indie UE5 demo of a horror game and when I entered dark rooms and turned on the light, it took almost half a second for the light to resolve, it was very jarring. 

[HMart]

9 minutes ago, Wellingtoncrab said:

More polys/geometry usually isn’t easier in the context of this game, where making content which runs at all is about producing clean sealing geometry and portalization.

For all its faults an illusion like POM, which derives detail while retaining minimally complex sealing geometry seems well suited to that and the workflow of most mappers.

Maybe hardware tessellation could achieve something similar with less of the drawbacks, but I am pretty happy with this so far.

Agree, I talked better about tess but for TDM POM is perfectly fine.  I expect Tess requires harder work at the engine geometry handling system, like a way to dynamically tesselate models and create LOD's. 

Edited December 15, 2024 by HMart

[Guest]

Yup, AFAIK, automatic hardware tessellation isn't great either. Typically, it creates way too many polygons for the difference it makes, which is again, already controllable via standard modelling workflow & LOD.

 

47 minutes ago, Wellingtoncrab said:

More polys/geometry usually isn’t easier in the context of this game, where making content which runs at all is about producing clean sealing geometry and portalization.

Did you do any experiments with existing LOD system? I didn't have anything that would be game-ready, but my initial experiments show that this engine could potentially handle massive polycount nicely, if other factors are kept at bay. Would have to get to semi-finished state with that, with some mission scripts, objective and other data set up to see how feasible it is.

Edited December 15, 2024 by Guest

[Wellingtoncrab]

@peter_spy The Pepsi POM challenge of what is more efficient is again maybe interesting, but for the workflow of this game does it even matter? Its already obvious POM is pepsi and real geometry is coke, but it presumes these would actually be interchangeable for most people making content for the game.

If I need a brick surface, I and 99% of mappers are going to drop a worldspawn brush and put a tiling brick texture on it using the inbuilt tools. I am not going to make a model of it. I am definitely not going to make LODs of it.

The use case of most of these POM materials is to be a familiar extension of the traditional brush/texturing/sealing workflow, not a replacement for models which have, and will always have, their place in the game.

[Guest]

Ah, you meant in the context of making assets for this game (or for any game, really). No argument there.

[stgatilov]

1 hour ago, Wellingtoncrab said:

If I need a brick surface, I and 99% of mappers are going to drop a worldspawn brush and put a tiling brick texture on it using the inbuilt tools. I am not going to make a model of it. I am definitely not going to make LODs of it.

On the other hand... what if you specify the heightmap and some parameters, and the engine generates real geometry out of it?

I'm interested in how geometry vs parallax mapping compete in terms of performance / quality on some typical scenario. Maybe we'd better thing about some auto-tessellation stuff.

[Wellingtoncrab]

2 minutes ago, stgatilov said:

On the other hand... what if you specify the heightmap and some parameters, and the engine generates real geometry out of it?

I'm interested in how geometry vs parallax mapping compete in terms of performance / quality on some typical scenario. Maybe we'd better thing about some auto-tessellation stuff.

Yes that would be very interesting. That’s why I mentioned tessellation as maybe an alternative with less downsides than POM when you are considering the actual brush/texturing workflow used when building levels for the game.

[Guest]

Default material:

With POM:

With some elements on geo:

These have no LOD, so that 23% GPU load looks a bit sus, maybe there was a hiccup there. All in all, performance stats do look similar.

As for POM, it does look pretty cool, 20 steps look good, even without much texture authoring on my end.

I sure wouldn't want to model these pebbles by hand

[stgatilov]

I'm not sure what is "GPU time %".
FPS on Vsync off is much more sane performance metric.

Did you check that quality is about the same when looking at the same distance?

How did you set texture size for parallax mapping?
Downsampling it might help performance.

[Guest]

It's the GPU load percentage. The quality and resolution settings are the same for all setups, I also use the same player start, just press crouch before taking screenshots. The heightmap size is the same size as base textures, which is 2k, but it's a 1-channel greyscale image.

When I started walking around, I got to a point where I got spikes with GPU load, like this:

I can't figure out why yet.

[Guest]

11 hours ago, peter_spy said:

I can't figure out why yet.

Ah, it's the Aniso x16 issue. I forgot to turn it down to x8. That's an issue going way back, by the way: sometimes when you look at a surface at acute angles, you get GPU load spikes, without any visible reason. I never had problems with other games going all the way to AF x16. Wonder what's with that in TDM.

Anyway, I tested it with uncap FPS and VSync off too. I've never heard my GPU fans being this loud
Default material:

Geometry:

POM:

[stgatilov]

May I get the testmaps with meshes and parallax?

[stgatilov]

I wrote a wiki page about the effect:
https://wiki.thedarkmod.com/index.php?title=Parallax_mapping

Please report if I forgot anything important...

 

[Baal]

@stgatilov

I tested the new _parallax materials in a map I'm working on. 

The machine I'm using has a relatively weak integrated GPU (it's an AMD Ryzen3). It runs most missions very well nowadays. But these materials have a big performance impact in still very basic scenes.

So I think there needs to be a way to disable the effect. Otherwise the minimum specs could rise quite a bit, if mappers start using them a lot. 

 

Edited January 20 by Baal

[nbohr1more]

20 minutes ago, Baal said:

@stgatilov

I tested the new _parallax materials in a map I'm working on. 

The machine I'm using has a relatively weak integrated GPU (it's an AMD Ryzen3). It runs most missions very well nowadays. But these materials have a big performance impact in still very basic scenes.

So I think there needs to be a way to disable the effect. Otherwise the minimum specs could rise quite a bit, if mappers start using them a lot. 

 

r_skipParallax 1

to disable.

You mean a gui option?

[Baal]

Thanks. I don't know how many players are impacted by this. But I would say, yes, an easy and obvious way to tweak it is necessary.

 

[SeriousToni]

On 1/20/2025 at 4:14 PM, Baal said:

@stgatilov

I tested the new _parallax materials in a map I'm working on. 

The machine I'm using has a relatively weak integrated GPU (it's an AMD Ryzen3). It runs most missions very well nowadays. But these materials have a big performance impact in still very basic scenes.

So I think there needs to be a way to disable the effect. Otherwise the minimum specs could rise quite a bit, if mappers start using them a lot. 

 

Is there already a map in which this effect is implemented? Wanted to see this very nice looking bumps myself without hassling around in any shader files and Dark Radiant

[datiswous]

On 1/4/2025 at 10:35 PM, stgatilov said:

Please report if I forgot anything important...

Well, could somebody add some screenshots to some of the parts of the article? Some screenshots would make it a bit more clear.

Edited February 26 by datiswous

[SeriousToni]

How does parallax perform on non straight, but curved walls?? Did anybody test it for seams?

[HMart]

9 minutes ago, SeriousToni said:

How does parallax perform on non straight, but curved walls?? Did anybody test it for seams?

Parallax Mapping can look odd in curved surfaces, if they have a strong curvature, about seams there's nothing people can do at the technique level (that doesn't destroy performance), is a limitation of the technique but is possible to hide them with other geometry.  Parallax Occlusion Mapping, can look very good but is a visual illusion and doesn't really tesselate a surface, so is best to have in mind its limitations and strengths to better take advantage of it.

[Wellingtoncrab]

23 minutes ago, SeriousToni said:

How does parallax perform on non straight, but curved walls?? Did anybody test it for seams?

How well it will “work” depends on the nature of the curve, the number/angle of the facets in the curve, the material, the texture scale of the material, etc but it typically looks fine imo and will probably often look more “seamless” than a 90 - especially on convex edges.

We tried to tune most of the materials added to core to at least look ok tiled across 90 degree convex edges.

[revelator]

nice looks great i must say, this engine is truely starting to take off.

volumetric skies would indeed be nice but i suspect if we go that way we might also need megamap support so big areas dont tank performance completely.

[Filizitas]

On 12/16/2024 at 8:13 PM, peter_spy said:

Ah, it's the Aniso x16 issue. I forgot to turn it down to x8. That's an issue going way back, by the way: sometimes when you look at a surface at acute angles, you get GPU load spikes, without any visible reason. I never had problems with other games going all the way to AF x16. Wonder what's with that in TDM.

Anyway, I tested it with uncap FPS and VSync off too. I've never heard my GPU fans being this loud
Default material:

Geometry:

POM:

it a shame really that we two never worked on something together 

[Guest]

We still can Although now I have some serious life matters to attend to, might need a couple of months before things settle down. After that I'll probably return to making stuff

[Arcturus]

{
blend parallaxmap
map textures/test_parallax/blocks_mixedsize02_multicolour_height
min -0.05
max 0.05 * sintable[time*0.5]
}