I’ve been having a rough time building the main branch without errors on my system. Don’t have issues with the release build so idk what’s going wrong, but I guess I’ll have to wait a bit longer to try out Strata. And yeah its wild how large the engine is right now. Its especially brutal when you have several versions installed at once haha! Oh well.
Yeah, the errors have been a lot worse recently, I think since they’re not releasing again until 5.1 (probably the fall unfortunately), they’re iterating a lot faster and tearing up other systems. Strata in particular is utterly changing the material and lighting pipeline, and because of how Nanite’s connected, mesh rendering as well. I’ve gotten it to build, but it’s caused random crashs, bugs and errors, enough that it’s too unstable to work with.
I checked on strata pretty recently, they fixed a bunch of lighting and translucency bugs, and it’s approaching the quality of the old path slowly but surely.
And yes, I can’t really have more than one build at this point: between the city sample’s 110Gb and _Main’s 160gb, a significant chunk of my hard drive is just UE5 experimentation.
Some good development news though:
- Nanite on masked materials appears to be working great, my last test supported over a million bush instances without breaking a sweat. Haven’t been able to try out WPO, but from the Changelogs support is very good right now.
- Lumen just got a slew of optimizations and is looking at around 15% better performance over 5.0.
- It says that translucency in RT reflections was improved, no idea if that means anything for lumen but excited by the prospect.
Basically, rendering is getting a lot more powerful and flexible for 5.1, and most of lumen and Nanite’s weird limitations will be gone. It’s going to be strange, especially being able to go all-out on animated, interactive foliage with basically zero repercussions.
1 Like
How was the culling on the leaf geometry? I think the future of nanite foliage will be in 2 LODs. Once for individual leaves for LOD0, and once the leaves would start getting culled a LOD1 with cards to preserve the faces that would otherwise be discarded and cause nanite foliage to look bare.
Not only would that allow for insane amounts of foliage instances, but a level of detail I don’t think anyone was expecting.
And what of nanite overdraw on the foliage?
That’s a very interesting question, and I can answer part of that: I used Unreal’s default bush in the starter content just as a baseline, and that asset is entirely carded, masked geometry. Since nothing was individually modeled, I can’t confirm your hypothesis of Nanite’s two LODs, but it’s certainly a very interesting one. The only problem I see is that one would either have to manually author two LODs and then hand them to Nanite to use intelligently, or nanite would need to figure out how to create them upon import (not impssible, just an interesting challenge).
Conviniently, I don’t think that’ll be a problem: in addition to the starter content bush, I also experimented with the plant from the Realistic Rendering demo, which features minor masked geometry but largely individually-modeled leaves. It also worked spectacularly, and with no visible geometry pop-in, fragmentation, or aliasing.
Overdraw is a tough question for me to answer, but largely for UI reasons: in the overdraw viewmode, everything has a white outline to make meshes easier to identify. With all of the masked geometry having that outline, pretty much the entire screen was white, making it hard to gauge how much overdraw was actually happening . From what I could gather however, overdraw was relatively reasonable, and it seems like Nanite culling has gotten much more fine-grained than it was in EA, closer to per-pixel.
And yes, the insane LOD and instance count will be very exciting. I love Nanite working for foliage, WPO and PDO (although why you need PDO support for nanite beyond legacy content vexes me). The only thing that bothers me is that I have no idea how nanite can work for these objects, or even how it can access the material information to begin with given that EA nanite depended on being able to remove material calculations from geometry rendering ones.
I came back to this old thread after rereading the discussion on methods to fake caustics. Now that Nvidia has released their NVRTX package for UE5, I actually think we’re looking at a very competent feature set now.
UE5 obviously supports lumen and nanite, which are super powerful on their own. The NVRTX version in addition has RTXDI, their unlimited shadowed lights solution, support for RT shadows with volumetric fog, a workaround to deal with RT shadow mismatches on Nanite geometry, and perhaps most interesting, their caustics system.
While I haven’t downloaded and tested it yet, I will soon: the notion of combining Lumen’s GI with real-time ray-traced fog, colored shadows and caustics sounds like a real visual treat, assuming they’re interconnected and optimized. I’m very excited.
Very interesting. I’ll have to check it out. I love what Nvidia’s been doing lately.
1 Like
You’ve named your inputs on the custom node, but you haven’t changed the name of the inputs in the HLSL code. Notice it says InputA and InputB in the node text but you have named them Lumen and Default.
Just for anyone who’s wondering and isn’t already aware: UE5_main for the past couple of months has supported multi-bounce lumen reflections, due for release in 5.3. If you have the processing power, it’s a solved problem, but the hacks as presented here are still extremely useful from a performance perspective.
1 Like
Ue5-lightning .pdf (18.5 MB)
Just in case, no I understand few things more