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

There should be really high quality captures of Keanu from Matrix Reloaded. But maybe Epic did not want to use those?
 
There should be really high quality captures of Keanu from Matrix Reloaded. But maybe Epic did not want to use those?

That's probably the scan with outdoor lighting they mentioned. It not that they didn't have scans, it's that they didn't have one that fit ideally into their pipeline.
 

My understanding was they brought in the coalition to help with Xbox, mostly because they wanted to make sure it was highly optimized for Series S, which has more difficult constraints. Doesn't seem that mysterious to me. Just bring in one of the most technically proficient Xbox developers that are highly involved in UE4/5 already.
 
@0:09 hum, that clipping never happened to me. I know the car crashes are very dynamic (or rarely the same) when shooting out the tires. So, it seems from that clipping, that transitional scenes are affected as well from the dynamic behavior of how one approaches these chases.

It's the same as any other real-time game with some scripted elements. The player car and the player avatar are probably not physics objects, so it clipped instead of colliding.
 
If you focus closely, you will see some geometry/Nanite LoD coming into focus, which from my perspective creates an illusion of shadows popping in.
Ah yes, I don't think any of those are due to shadowing primarily. It's hard to tell exactly for each case without further investigation, but most of those are likely HLOD transitions as milk noted; if you turn on the Nanite Instances view you might get some additional information although given how these buildings are constructed even instances view can look like triangle view at a distance (which in itself is sort of impressive IMO...).

I don't doubt most of those are pretty easy to solve with a bit more polish, but there was only so much time and resources for the large scope of the sandbox world.

There's a few other little issues scattered around like Z-fighting of extra coincident planes on roofs and some weird procedural generation quirks. There's spots that exceed AA's ability to entirely eliminate the aliasing/moire as well. Overall I think it's still a fairly stable presentation compared to past games though, and Nanite and VSMs are a decent chunk of that.

Is that true for open world games? For the dynamic sun? I assumed these are tweaked once for the most robust solution for all possible scenarios, and then begrudgingly accept the artifacts that can't be avoided. No? Are there ways to tweak it more beyond that?
I was referring mostly to the chase sequence where there were no settings differences or tweaks done to the directional light shadowing, even in very close up shots with narrow FOVs, etc. Obviously in a real game it's very common to highly tweak those sorts of on-rails segments.

That said, there's still some cases that get tweaked in open world games. CSM distances are usually pulled in and some sort of "far shadows" solution is employed (usually baked or at least heavily cached or something like DF shadows). In this case the virtual shadow map is handling shadows all the way from right up close to the entire city/horizon. Sunset/sunrise normally needs special treatment (read: avoiding it as much as possible :p) due to low angle sun and biasing issues.

VSMs are not entirely immune to low light angle issues either, but it's a much lower light angle than most games could get away with in this demo and still holds up fairly well IMO. If you swap to CSMs there are significant biasing artifacts all over the place, resolution and shadow distance limitations are quite apparent. There are of course new considerations with VSMs around minimizing invalidations and picking your moving geometry a bit more carefully, but I think it's fair to say even for an open world game it is much more hands-off for artists than previous solutions have been.

Why wouldn't they use Nanite LOD changes in those buildings? Being completely static geometry it seems like it's the perfect scenario for Nanite.
There's still going to be a point where even *instances* are too small and need to be collapsed in a scene this big with so much detail. Nanite can push that out much further than other solutions, but as I mentioned there's also other boring engine and CPU overhead reasons why HLOD still exists for these large scenes (albeit simplified and doesn't need to handle as much heavy-lifting with Nanite).

Hey Andrew, could we have some info regarding the difference in performance between sw Lumen and hw Lumen?
I'm not super familiar with the details there and wouldn't want to speak for the Lumen folks on that even if I were.
 
Ah yes, I don't think any of those are due to shadowing primarily. It's hard to tell exactly for each case without further investigation, but most of those are likely HLOD transitions as milk noted; if you turn on the Nanite Instances view you might get some additional information although given how these buildings are constructed even instances view can look like triangle view at a distance (which in itself is sort of impressive IMO...).

I don't doubt most of those are pretty easy to solve with a bit more polish, but there was only so much time and resources for the large scope of the sandbox world.

There's a few other little issues scattered around like Z-fighting of extra coincident planes on roofs and some weird procedural generation quirks. There's spots that exceed AA's ability to entirely eliminate the aliasing/moire as well. Overall I think it's still a fairly stable presentation compared to past games though, and Nanite and VSMs are a decent chunk of that.


I was referring mostly to the chase sequence where there were no settings differences or tweaks done to the directional light shadowing, even in very close up shots with narrow FOVs, etc. Obviously in a real game it's very common to highly tweak those sorts of on-rails segments.

That said, there's still some cases that get tweaked in open world games. CSM distances are usually pulled in and some sort of "far shadows" solution is employed (usually baked or at least heavily cached or something like DF shadows). In this case the virtual shadow map is handling shadows all the way from right up close to the entire city/horizon. Sunset/sunrise normally needs special treatment (read: avoiding it as much as possible :p) due to low angle sun and biasing issues.

VSMs are not entirely immune to low light angle issues either, but it's a much lower light angle than most games could get away with in this demo and still holds up fairly well IMO. If you swap to CSMs there are significant biasing artifacts all over the place, resolution and shadow distance limitations are quite apparent. There are of course new considerations with VSMs around minimizing invalidations and picking your moving geometry a bit more carefully, but I think it's fair to say even for an open world game it is much more hands-off for artists than previous solutions have been.


There's still going to be a point where even *instances* are too small and need to be collapsed in a scene this big with so much detail. Nanite can push that out much further than other solutions, but as I mentioned there's also other boring engine and CPU overhead reasons why HLOD still exists for these large scenes (albeit simplified and doesn't need to handle as much heavy-lifting with Nanite).


I'm not super familiar with the details there and wouldn't want to speak for the Lumen folks on that even if I were.

Dude, thank you so much for all the transparency. As it is right now, does HLOD have multiple levels too, or just one? And do theses colapsed models also get to be processed through Nanite? Are you guys researching ways to also make that part of it as automated as possible and a integral part of nanite itself?
 

I experimented creating traffic for one hour and blocked multiple routes.
Some strange observations.
Some traffic jams seemed to reset/remove some cars.
The big traffic jam was increasing, but at some points, the stopped traffic started moving slowly, which gave an indication that some cars were disappearing.
Another route I blocked was collecting cars, but when I revisited the length of the traffic jam was reduced.
Same for another smaller one.
Also if I damage a car and leave it to create a traffic jam, go away and come back, the car is no longer damaged and is replaced with its original nanite geometry.
Btw you can see some pop in and out with the traffic jam cars on the highway bridge. Very large distances disappear the cars and appear as you approach
 
There's always a hardware limit for sure.
But compared to other openworld games, there must be like 100 times the number of cars on screen ?

This engine would be great for a Superman / Man of Steel game. Even with this small sample, it still feels like a huge living Metropolis while flying around. The amount of geometry, life-like vertically structures, insane amount of traffic/NPCs, high frequency details/textures, and everything else that is going on, it's just ungodly at this point. I really can't believe PS5 and XBSX/S are actually capable of handling such an engine. I really want this demo on PC, so I can crank up the settings even further.
 
For a MoS game, you'd need all the buildings to be trashable! Better suited to less destructive heroes, methinks. ;)

Well, have you seen all the destruction caused by Marvel characters and DC's Batman have done on screen? Yet, they have games.

But, I get what you're aiming at though...
 
On the highways, it seems the semi-trailers never use the left lane?
 
On the highways, it seems the semi-trailers never use the left lane?

In certain states, left lane usage by 18-wheelers/semis/large trucks is prohibited. I guess that applies to the Matrix metropolis... :yep2:
 
Last edited:
Back
Top