Unreal Engine 5, [UE5 Developer Availability 2022-04-05]

[Shifty Geezer]

Unfortunately not, real life snow tends to look pretty weird and very obviously crunchy/fluffy, to the point where you can see the individual crystals from a decent distance away, the sparkle shaders we have now are a fantastically poor approximation. Not to mention how poor the heightfield hacks when walking through snow look.

There's been entire papers written on it, and researchers asked Disney for their snow physics sim from Frozen because it was the first accurate one they'd seen.

I don't really understand this argument. We don't model anything realistically in realtime graphics. In shading skin, we don't model light transmission and reflection through multiple layers of different cells and oils and varying water content and blood vessels. In shading foliage, we don't model light transport through layers of cellulose wall and pigments and the air in between to get accurate light scattering. In creating hair, we don't shade the light transport through hundreds of stacked, hollow, pigment-filled, oil-covered fibres, reflecting and refracting as happens in real life. We approximate everything. Peach fuzz? Don't model hundreds of tiny hairs and shade them at the sub pixel level - add a rim-light approximation! I don't see anything about snow that makes it harder than anything else. If snow shaders are weaker than other materials, I'd say that's a lack of investment because it's a niche material, in contrast to human skin that has a lot of effort coming up with usable approximations.

As for the physics, I thought this was a statement about the static snow in the scene. which wasn't being interacted with. Accurate snow simulation, like everything else, will be approximated somehow because it's impossible to simulate accurately. Just as lots and lots of work has been done making water realistic (which it isn't particularly, but it looks pretty good in games) the more games use snow, the more it'll be modelled. I expect it'll end up with a generic ML approximator along with other similar mixed-behaviour materials like sand.

 

[Frenetic Pony]

I don't really understand this argument. We don't model anything realistically in realtime graphics. In shading skin, we don't model light transmission and reflection through multiple layers of different cells and oils and varying water content and blood vessels. In shading foliage, we don't model light transport through layers of cellulose wall and pigments and the air in between to get accurate light scattering. In creating hair, we don't shade the light transport through hundreds of stacked, hollow, pigment-filled, oil-covered fibres, reflecting and refracting as happens in real life. We approximate everything. Peach fuzz? Don't model hundreds of tiny hairs and shade them at the sub pixel level - add a rim-light approximation! I don't see anything about snow that makes it harder than anything else. If snow shaders are weaker than other materials, I'd say that's a lack of investment because it's a niche material, in contrast to human skin that has a lot of effort coming up with usable approximations.

As for the physics, I thought this was a statement about the static snow in the scene. which wasn't being interacted with. Accurate snow simulation, like everything else, will be approximated somehow because it's impossible to simulate accurately. Just as lots and lots of work has been done making water realistic (which it isn't particularly, but it looks pretty good in games) the more games use snow, the more it'll be modelled. I expect it'll end up with a generic ML approximator along with other similar mixed-behaviour materials like sand.

Because here's what real snow looks like:



And here's what game snow looks like even on the highest end today:



Comparatively, the game snow is obviously very flat and smooth all around, the shiny specks look pasted on top rather than coming from the snow itself, etc. Getting to the hyper detail of the real life stuff is a major challenge.

 

[Frenetic Pony]

Addendum: For reference correct lightbounce in between snow should be achievable in realtime now. The same energy preserving RT subsurface scattering shown in The Callisto Protocol and UE5 should be able to well approximate how light goes through packed snow.

The major challenge for realtime is getting the hyper detail of the surface of snow when you get close. The crystals are small enough that nanite or something like tessellation would have a tough time; and getting the sparkles to actually match up to surface is a challenge even for offline rendering. There's been a lot of research there in stable "sparkle rendering", so the sparkles you see from a distance match up to individual bits close up. As there's only research papers on that for offline rendering it's going to be much harder to do that in realtime, rather than having a cheap-ish arbitrary layer of sparkles just kind of pasted on top.

 

[neckthrough]

Because here's what real snow looks like:
<snip>
Comparatively, the game snow is obviously very flat and smooth all around, the shiny specks look pasted on top rather than coming from the snow itself, etc. Getting to the hyper detail of the real life stuff is a major challenge.

To convince @Shifty Geezer it's not enough to show that there's a difference between real and simulated snow. He accepts that. You'll have to demonstrate that this difference is notably worse for snow than for other materials.

[ For the record I don't have any skin in this game. I'm just serving as an interested observer and conversation facilitator ]

 

[PSman1700]

Never thought about how hard it is to render realistic snow lol. Should be the same for some types of sand? I thought SSX Tricky did good snow on the OG xbox back in 2001 using bump mapping. Not really impressive nowadays

 

[trinibwoy]

Because here's what real snow looks like:

View attachment 7791

And here's what game snow looks like even on the highest end today:

View attachment 7792

Comparatively, the game snow is obviously very flat and smooth all around, the shiny specks look pasted on top rather than coming from the snow itself, etc. Getting to the hyper detail of the real life stuff is a major challenge.

You would need pixel sized geometry and some whacked out shaders to replicate that look. Making it deformable is another level of impossible.

 

[Shifty Geezer]

Because here's what real snow looks like:

The scene above that prompted the mention of snow does not look at snow that closely. It looks at snow at this distance and depth...



You said the snow wasn't convincing and it'd need crazy work to make look as good as other parts of the render. "Can confirm from experience that this matches real life experience pretty solidly. Well, except for the snow." I say it doesn't need that much work to get that snow looking better, it just needs a better shader. If the criticism of UE5 and games in general is that they aren't recreating deep snow like this...

You're right, that'd need some fancy work, but I don't know why you are referencing that model in response to a light dusting of snow in the UE5 demo you responded to and arguing with me that my response to how to make that smattering-of-snow look good via shaders fails because it wouldn't solve deep snow simulation.

As for producing an accurate snow simulator, yep, it's got a long way to go, but that's true of many other simulations. Clothing is still crap, with lots of baked pieces that warp instead of slide over each other. We're only just getting good face and hair simulation and that's with insane amounts of work because characters are core to a lot of games. I still don't see that snow is inherently harder - it's just lower priority as it's more niche. And it still wouldn't use 50 bounces through a complex BSDF, just as we don't model that level of realism in tracing through skin etc. Unless you can present an argument that snow rendering is more complex and sophisticated than skin rendering, you won't win me over with your pro-snow-complexity argument.

My prediction - the day we have a good approximation of snow in games, it'll be via some approximation/mathematical model, almost certainly an ML solver, and no-one anywhere will be trying to model the details you've mentioned.

 

[Silent_Buddha]

Hell, we're only now getting trees with actual leaves over long distances despite hardware support for instancing on GPUs being a thing for about 2 decades now. Obviously because there's more involved than just instancing all of those millions of leaves in a large scene.

And that's something that is far more visible and far more prevalent in games than snow. So, it's understandable that something like snow (especially deformable snow) is using very rough approximations.

Regards,
SB

 

[Scott_Arm]

I'm very familiar with snow. If you're out in an area where the snow is undisturbed, especially up in mountains where the snow is very powdery, that Red Dead Redemption 2 screen is actually a reasonable approximation.

 

[Shifty Geezer]

I'm very familiar with snow. If you're out in an area where the snow is undisturbed, especially up in mountains where the snow is very powdery, that Red Dead Redemption 2 screen is actually a reasonable approximation.

It actually snowed here in the SE of England this week. Looking out my window, the snow looks pretty like that UE5 woodland example. Not even any sparkles - just a sheet of uniform white where it's settled. Of course, there are different types of snow that those future snow-simulation engineers will have to integrate into their models.

 

[Scott_Arm]

It actually snowed here in the SE of England this week. Looking out my window, the snow looks pretty like that UE5 woodland example. Not even any sparkles - just a sheet of uniform white where it's settled. Of course, there are different types of snow that those future snow-simulation engineers will have to integrate into their models.

View attachment 7814

At least 50% of gpu power should be spent on rendering snow.

 

[Karamazov]

We should have a "best snow in a game" thread

 

[Shifty Geezer]

We should have a "best snow in a game" thread

Yup. Used to be water. That's largely passé now. Should move on to frozen water as a new benchmark. Can couple that with water vapour when the sun hits a dark surface, melts the snow and evaporates the water as visible water vapour. At least that one is a solved challenge now via 3D compute-driven volumetrics, even if no game is spending the cycles on that.

 

[Silent_Buddha]

Yup. Used to be water. That's largely passé now. Should move on to frozen water as a new benchmark. Can couple that with water vapour when the sun hits a dark surface, melts the snow and evaporates the water as visible water vapour. At least that one is a solved challenge now via 3D compute-driven volumetrics, even if no game is spending the cycles on that.

Considering I've been shoveling snow off and on for the past 4 weeks now, we need to also make sure we properly model "wet" snow (at the extreme it's almost slushy), crunchy "frozen" snow, powdery snow, snow frozen into rock hard chunks, etc. Although I wouldn't be at all sad if someone (God maybe?) banned wet snow and erased it from existence, I hate shoveling "wet" snow.

Regards,
SB

 

[BRiT]

More importantly, don't eat the yellow snow...

 

[dobwal]

All we need is a convincing appromixation. Nothing more or nothing less.

Or what? Devs are going to need to code temperature into the game and then use ray tracing to calculate the varying temperatures throughout a scene? Temperature of any surface will be higher in direct sunlight versus in the shade. Ain’t nothing worse than a storm to drop a couple feet of snow only to be followed by a bright, sunny and relatively warm day then drop back into freezing temperatures overnight. You know driving is going to be treacherous the next morning. Why do I need that in a game?

All most need is to have snow good enough not to break immersion. Like not bothering and letting characters just clip through knee deep snow.

 

[Frenetic Pony]

All we need is a convincing appromixation. Nothing more or nothing less.

Or what? Devs are going to need to code temperature into the game and then use ray tracing to calculate the varying temperatures throughout a scene? Temperature of any surface will be higher in direct sunlight versus in the shade. Ain’t nothing worse than a storm to drop a couple feet of snow only to be followed by a bright, sunny and relatively warm day then drop back into freezing temperatures overnight. You know driving is going to be treacherous the next morning. Why do I need that in a game?

All most need is to have snow good enough not to break immersion. Like not bothering and letting characters just clip through knee deep snow.

I mean, you could say that literally about every piece of gamedev. ASCII is a good enough approximation of visualization, why do you need a tileset of pixelart for Dwarf Fortress? This is literally a thread about high end rendering in video games. UE5 already goes super into detail on hair and skin and now layered materials about how water can drip down over the clearcoat over the paint layer on a car, of course this is relevant. Snow in games looks kinda shit, it could look better, cool.

 

[dobwal]

I mean, you could say that literally about every piece of gamedev. ASCII is a good enough approximation of visualization, why do you need a tileset of pixelart for Dwarf Fortress? This is literally a thread about high end rendering in video games. UE5 already goes super into detail on hair and skin and now layered materials about how water can drip down over the clearcoat over the paint layer on a car, of course this is relevant. Snow in games looks kinda shit, it could look better, cool.

What is good enough is determined by the masses not any one individual. Cars, skin and hair appear in most games at a relatively high rate so it is easy to see why time and effort is devoted to those visuals. There are whole games devoted to automobiles. How many to snow?

In all the time you have been here, how many discussions have you had about how bad snow looks in games in general?

It would nice if someone made an effort and pulled off something worth noting but it’s hardly a priority.

 

[chris1515]

 

[davis.anthony]

We should have a "best snow in a game" thread

Spintires Snowrunner

/end