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

It depends of the TAA implementation.
As pointed out, depends on TAA implementation.
Most are bad anyway.
And I'll take jaggies at 32" 1440p any day over DLSS* or bad
Jaggies are bad, no matter how you slice it, and in most engines and games you can't disable TAA, so DLSS 2 is the best option right now, especially in motion, where the ringing artifacts are least noticeable anyway.
 
as you said most implementations of TAA @ 30hz is bad, but as you scale up the frame rate not only do the TAA ghosting artifacts become less noticeable they are also considerable smother, so something like 120hz would be the killer feature that TAA games need next gen.
 
Are there any mainstream upscaling implementations which use triangle ID? Using Triangle IDs could help solve a lot of problems, having access to G-buffer lighting information even more.
 
Last edited:
Game File Sizes May Skyrocket with Unreal Engine 5’s Nanite, Says Developer
July 18, 2020
The issue with game file sizes may be about to become even more prevalent with the upcoming Unreal Engine 5, which promises astounding advancements in visual quality that could however translate into even larger game file sizes. When we asked Kitatus and Friends CEO and Lead Programmer Ryan Shah what he thinks of the Unreal Engine 5 as an existing Unreal Engine developer, he told us - after plenty of praises for Epic - that could indeed be a problem.
...
The problem is inevitably going to be compounded by the fact that storage solutions may be getting faster with the PlayStation 5 and Xbox Series X consoles, but they're certainly not getting larger. Sony's console has 825 GB of space, while Microsoft's has a bit more room with 1 TB.

How many Unreal Engine 5 games will fit in the respective SSDs? We don't know yet, but Epic is aware of the potential issue regarding download sizes, so we can hope they'll be able to find a workaround by the time it ships next year.
https://wccftech.com/game-file-sizes-may-skyrocket-with-unreal-engine-5s-nanite-says-developer/
 
Are there any mainstream upscaling implementations which use triangle ID? Using Triangle IDs could help solve a lot of problems, having access to G-buffer lighting information even more.

Nothing, I don't think a deferred texturing (hand in hand with triangle id buffers) rendered game has shipped yet. Wouldn't be surprised to see it starting next gen, as it requires some fundamental changes to the rendering pipeline. And that means any 4/5 year long game, IE one that started work back when all the triangle ID/deferred texturing stuff started showing up would be juuuust about to hit.
 
Initial tweet leading to that one --

Great to see UE5 using virtual shadow mapping for their Nanite tech. Their shadows look awesome. Screenshot below from my SIGGRAPH 2015 presentation. Too bad I only had 30 minute slot. Virtual shadow mapping was compressed to one slide :(

In this example we had 1:1 texel : pixel shadows everywhere. And it was still much faster to render than a cascaded shadow map.
 
I'm really excited to learn more about this (and try it out!) The promise of what they're selling is obvious but I think it has much more potential than many non artists are recognizing.

I don't have the time to read through all 50 pages of the thread, has there been any particularly good speculation on the technique?

My main point of concern is whether they choose the desert environment out of performance limitations, rather than art expediency -- the real potential of unlimited polygon count is things that are traditionally done with alpha transparency, imo -- dense foliage, detailed ground cover and debris, vines, reeds, etc. In the demo we saw few thin or overlapping details, just the banners and statues. it is speculation (more of a wild guess) but I'm worried there might be some cost to rendering surfaces that overlap or are in close proximity (like with many acceleration structures for raytracers)
 
My understanding is transparency is accomplished using the graphics pipeline rather than compute. I haven't seen any speculation about objects in close proximity.
 
I'm really excited to learn more about this (and try it out!) The promise of what they're selling is obvious but I think it has much more potential than many non artists are recognizing.

I don't have the time to read through all 50 pages of the thread, has there been any particularly good speculation on the technique?

My main point of concern is whether they choose the desert environment out of performance limitations, rather than art expediency -- the real potential of unlimited polygon count is things that are traditionally done with alpha transparency, imo -- dense foliage, detailed ground cover and debris, vines, reeds, etc. In the demo we saw few thin or overlapping details, just the banners and statues. it is speculation (more of a wild guess) but I'm worried there might be some cost to rendering surfaces that overlap or are in close proximity (like with many acceleration structures for raytracers)

Nanite isn't suitable for foliage or skinned objects. You can mix and match Nanite and regular geometry in a scene though. "Make an object Nanite" is an editor tick box.

I thought geometry images seems like as good a speculation as any from the thread. Ties into 1m polys being around the stated same as 4k texture in size. Brian Karis blogged about them a decade ago, so maybe it was just the springboard of an idea.

http://graphicrants.blogspot.com/2009/01/virtual-geometry-images.html
 
Nanite isn't suitable for foliage or skinned objects. You can mix and match Nanite and regular geometry in a scene though. "Make an object Nanite" is an editor tick box.

I thought geometry images seems like as good a speculation as any from the thread. Ties into 1m polys being around the stated same as 4k texture in size. Brian Karis blogged about them a decade ago, so maybe it was just the springboard of an idea.

http://graphicrants.blogspot.com/2009/01/virtual-geometry-images.html

Oops, yeah, foliage was a bad example -- i knew about skinned meshes, but my brain wasnt working. I mean anything that's thin/covers a lot of the screen/overlaps a lot of objects in depth -- metal fences are a rigid example that springs to mind. Except for the statues, the environment in the demo was all basically flat surfaces with surface detail and not a lot inbetween, which makes me wonder if there's some hidden cost or limitation there.

Geometry images are cool, thanks for the tip
 
Oops, yeah, foliage was a bad example -- i knew about skinned meshes, but my brain wasnt working. I mean anything that's thin/covers a lot of the screen/overlaps a lot of objects in depth -- metal fences are a rigid example that springs to mind. Except for the statues, the environment in the demo was all basically flat surfaces with surface detail and not a lot inbetween, which makes me wonder if there's some hidden cost or limitation there.

Geometry images are cool, thanks for the tip

Things with many small objects that become porous at a distance are not a good fit for Nanite. Not sure if a fence fit that category. They did have scaffolds in the demo at various points.

They think that Nanite covers >90% of geometry though.

Screenshot_2020-11-23-19-54-58-52.jpg
 
What are traditional techniques used in games for smooth LOD transition between dense foliage and volumetric approximations? I don't believe there is any tradition to speak of. Plenty of old research, but not much tradition.
 
What are traditional techniques used in games for smooth LOD transition between dense foliage and volumetric approximations? I don't believe there is any tradition to speak of. Plenty of old research, but not much tradition.

In this instance, it'd be just whatever you'd use in UE4.xx.
 
Back
Top