Unreal Engine 5, [UE5 Developer Availability 2022-04-05]
- Thread starter mpg1
- Start date
Silent_Buddha
Legend
Since people are still talking about it. 
A small example of swimming character meshes in Messiah. The scientist on the right. Courtesy of Digital Foundry.
Should be timestamped at 24:16.
Regards,
SB
A small example of swimming character meshes in Messiah. The scientist on the right.
Courtesy of Digital Foundry.Should be timestamped at 24:16.
Regards,
SB
Old good nyquist sampling theorem fits well to displaced surfaces as well.Since people are still talking about it.
A small example of swimming character meshes in Messiah. The scientist on the right.
Should be timestamped at 24:16.
Regards,
SB
Characters were few cylinders with tesselation and hole filling polygons in between.
Pretty sure they didn't double amount of displaced points each LoD level, so displacement sampling locations could move around full resolution heightmap.
With properly selected mipmapping and tesselation sampling forced to keep sample point locations, it might become quite stable.
Jak and Daxter is a great example of how to do smooth LoD morphing right. Only used for the environment, though...
Most env. assets have 2 or 3 LoD levels wich smoothly blen from one and another. Its specially easy to see when playing at higher resolutions than intended through emulation.
Smooth tesselation morphing is done by doubling the tess factor all at once, and ajusting the displacement smoothly.
For some reason, the standard DX11 aproach was not that. It smoothly moved the new vertices tangentially through the surface out of a "mother" vert, all while sampling the heightmap along the way. That aproach only maximized the "swimming". Missed oportunity...
Yup.
Seems like DX11 tesselation got a lot wrong, haven't really heard anything good about it.
Nothing about hardware tessellation prevents tessellating an entire mesh at once. Doing that is not performant, and also has little benefit over just sending a higher res mesh to the gpu in the first place. The main advantage of hardware tesselation is dynamically changing tessellation per patch.
Just stumbled across this Gamewatcher article again (from 22nd Nov). Some games already out now, but plenty more to wet the appetite for next year
www.gamewatcher.com
UNREAL ENGINE 5 GAMES LIST - UPCOMING UE5 TITLES
Unreal Engine 5 Games List - Upcoming UE5 Titles | GameWatcher
Unreal Engine 5 has been revealed to the world and we already have a list of games confirmed to use it.
www.gamewatcher.com
Will Nanite allow for universally indentical detail level? EDIT: In other words: zoom in and out with no change in graphical detail.
For example, I'd like to play a game where I can turn from a giant into an ant. Imagine a gameplay situation where you are stomping around as a two-legged giant creature in a densely underbrushed forest. With the press of a key you'd begin to transform into an ant. I say 'begin' because the transition would not be instant but smooth and rather slow, so you'd be able to clearly see everything during the miniaturizing and enlarging process.
For example, I'd like to play a game where I can turn from a giant into an ant. Imagine a gameplay situation where you are stomping around as a two-legged giant creature in a densely underbrushed forest. With the press of a key you'd begin to transform into an ant. I say 'begin' because the transition would not be instant but smooth and rather slow, so you'd be able to clearly see everything during the miniaturizing and enlarging process.
Last edited:
Thats not what I meant. One still wants adaptive tesselation based on patch size on screen. My point is the way the transitions were handled in the standard DX11 aproach made swimming worse.
cheapchips
Veteran
Antgiant feels like a bit of niche genre.
The theme is irrelevant to the question. It was just an example.
cheapchips
Veteran
Maybe you could pitch this for Knack 3?
The limiting factor there will be content requirements. You'd need the whole world modelled at a scale for a giant also being modelled in the detail for the ant scale. Short of some incredible online streaming tech, you couldn't fit enough data in the local storage for such graphical density, plus the costs to produce are likely prohibitive. Even using photogrammetry, take a street for example, you'll need to photo the whole thing an inch at a time for ant scale! Capturing an entire city, inside and out, at the detail level suited for an ant, would take too much time and effort and, even if that were automated, too much data.
Keeping the dream alive though, if a network title were made with a subscription fee and some robotic sampling system and some godly fast internet connection, it could be achieved.
Last edited:
I looked at the Unreal Engine 5 version of Fortnite and I am very satisfied with the TSR upscaling, which runs together with dynamic resolution. I think it raises the low native resolution to a great image quality. In general, what are the experiences and opinions about TSR?
DavidGraham
Veteran
Interesting differences between offline RT light baking and Real Time Lumen.
cheapchips
Veteran
I'm curious as to whether a minimal / very white indoor environment is an easy or challenging use case for Lumen.
Last edited:
It's OK. There was already DRS before. The problem is it's hard to compare to others solution. But I immediatly noticed the new Fortnite was running at a lower resolution with more aliasing. So it's a trade-off, less resolution in exchange of RT reflections, GI and higher poly count. In the end the image fidelity obviously greatly improved as the loss of resolution is not a big deal because of high diminishing returns.
But there is an important downgrade though: the lack of self-reflections in mirrors (the base PS4 version still has it!) ...yes I am still not happy about that.
I'm not sure about this specific case, but in general arch vis scenes with clean lines and light colors are more challenging as a significant portion of the lighting comes from bounced (often multi-bounced) light and there are no diffuse textures or complex geometry/direct lighting to hide noise.
Of course they are simpler from a geometric complexity perspective though, and thus often are a good fit for triangle RT.
Frenetic Pony
Veteran
Still a decent amount of energy loss in Lumen. Distributing the surface cache probes more evenly (blue noise evangelism) would probably help, but the biggest thing I can think of is feedback RESTIR GI sampling bias.
Similar threads
