I honestly figured as much, at this point. With four major public revisions of lumen (E.A-5.3), I struggle to think of must-have featuresets that aren’t supported in some way. With larger codebases and interdependent systems, it makes sense that forward momentum would become slower. I know many people will be more than happy with bug fixes and performance improvements, and the new features can arrive when they may. I really appreciate you being on the forums and letting the community know what to expect, thank you.
1 Like
DLSS 3.5? In everything that I’ve seen, it looks like temporal resolve is significantly improved compared to, as NVIDIA puts it, hand-tuned denoisers. It certainly looks better than anything else I’ve ever seen.
Daniel Wright stated that bugfixes and performance improvements are currently top of the priority stack for lumen. How lumen can function without temporal accumulation is a question for the coding oracles.
Speaking from the AAA devs I’ve talked to who have shipped titles, I don’t think ‘common sense’ and ‘game algorithm’ ever go together in the same sentence. Games are built out of hacks and kludges and cringeworthy last-minute fixes to ship. They are anything but elegant, but they are what games are as an art form. If you don’t want to use the newest technologies, that’s more than fair, but the lumen team is doing the best they can, and I think it’s particularly kind of Daniel Wright to visit the forums and let us know what the progress looks like.
Unreal Engine is now a cinematic and movie creation tool as much as a game engine, and Epic has a responsibility to both userbases. I understand that you’re upset with the limitations of the technology you’re facing, but the rest of us are as well. Let’s just do our best to be polite, to the people and the forum that lets us do the work that we do.
4 Likes
I was actually thinking less about foliage and more for materials like thin fabrics where you get very noticeable shadows you can use for shadow puppet effects or just to give mood to a scene, like tents, shoji doors. To some extent I suppose it can be worked around with emissives.
Anyway, thanks for taking the time to respond. I can appreciate that its a niche issue and there are much more pressing concerns to consider 
On the bug note as well:
Car paint in the city sample glows white when reflections are enabled with 2+ bounces and are out of screen-space.
!
The color appears completely white and doesn’t appear to be receiving any lighting at all, even as the wheel is clearly receiving direct lighting from that point light. I’m guessing some issue with the car paint shading model?
Importantly however, it’s not all cars, but I can’t figure out the rhyme or reason as to which ones seem to be bugging out. Building windows also seem to be failing to reflect in reflections, and car windows seem broken in their own ways.
2 Likes
DLSS 3.5? In everything that I’ve seen, it looks like temporal resolve is significantly improved compared to
Did you watch videos not sponsored by Nvidia, real gamers, or test done in non-“4k”?
RR hates motion completely because it when something moves, RR drags slime across that movement.
performance improvements are currently top of the priority stack for lumen. How lumen can function without temporal accumulation is a question for the coding oracles.
Awesome to hear the first part. But I’m not talking about dissing about Lumen slow accumulation due to a temporal design.
I’m talking about the flickering not being fixed without adding TAA or some other temporal methods like TSR or DLAA.
Currently two pathways are here-Crank up ScreenProbeGather.Temporal.MaxFramesAccumulatedto 25 to stop the flickering with SMAA, FXAA, no AA etc. But then you get light smearing when the camera pans
This is especially bad in third person games, characters get their face smeared when you pan the camera, or run etc.
Or you can lower the Cvar to 10 to stop light smearing and then rely on TAA,DLAA etc to hide lumens flickering.
I don’t care what temporal AA or pixel upscaler like DLSS, it ruins motion/gameplay.
Unless you are playing at 4k, which is not a realistic resolution since LOT of players with even $600+ systems can’t achieve 4k or even at 4k60fps.
Sharpeners are ugly and don’t fix motion smearing.
This is the oxymoron of modern game graphics.
[/details]
I don’t think ‘common sense’ and ‘game algorithm’ ever go together—
but the lumen team is doing the best they can, and I think it’s particularly kind of Daniel Wright to visit the forums and let us know what the progress looks like.—
Maybe I didn’t state that right but this post does ←
There is no point in making a game look amazing if upscales from 720p and looks like crap in motion. There also seems to be a lack scene specific optimization.
My message to the Lumen team: Thank you for Lumen.
My distaste is for the temporal AA obsession and blurry games. Lumen is not apart of that problem except that it pretty much forces developers to use TAA to fix Lumen current issues…rather than the issues in Lumen be fixed and not slapping on TAA/DLAA etc.
and Epic has a responsibility to both userbases. I understand that you’re upset with the limitations of the technology you’re facing
Keep it very separate then. Make a Virtual production mode/preset. Because the majority of game studios’ are ran by morons and lairs. I’m not worried about the performance of my game at this point.
You should know I’m concerned with every other game not produced by me at this point.
I’m not using Nanite now that we clear test vs blindly listening the the false Nanite-hyped documentation.
The problems I present are common sense. I don’t mean “common sense” as an insult. It’s simply my way of expressing: View the entire situation as a whole, including the major challenges and issues. Including the current issues of the so called “fixes.”
A game is meant to dynamically move. Cinematics can render in slow motion and TAA/DLAA etc isn’t going to hurt it.
Speaking of the Nanite test. The Lumen team needs to not revolve fixes, quality, and performance improvements around Nanite meshes only.
Do not force people to work with a unperformant renderer.
Optimized LODs are still the top dogs in performance and a better performing background will only make Lumen’s cost less concerning for 16ms budget games.
I can’t see anything bad in the comparison above. I’m not even sure what am I suppose to see. The image from the right shows bad aliasing and exaggerated sharpness, while the image from the left is much nicer with decent anti-aliasing and no smearing or blurring.
Maybe people don’t see what you see and that’s why there’s a lack of interest.
I can’t see anything bad in the comparison above. I’m not even sure what am I suppose to see.
What is wrong with your eyes? Tekken 8 TAA motion test - Imgsli
How do you not see how much the TAA smears and blurs the character, the fence, the ground, the effects, the words, the entire background? You have FXAA or SMAA to fix the Aliasing, I shouldn’t have to say that.
A screenshot can’t even show how bad it look in motion.
Here is entire test done on motion->Tekken 8 medium motion test(Jin B+4) - Imgsli
TAA is worst thing ever happened to games. For dozens of more comparisons go here.
the image from the left is much nicer with decent anti-aliasing and no smearing or blurring.
Are you kidding me? You’re joking right? The left in real gameplay is unplayable. It’s garbage. And I mean 100% garbage. That’s not even the default TAA in UE5 which blends more past frames/vaseline.
Maybe people don’t see what you see and that’s why there’s a lack of interest.
Yes… that is major issue, showing how bad TAA can be. TAA hates motion and that subreddit shows it better than I could.
In case the Lumen team can’t see it, many other people can and hate it. They hate it so much people will drop the game entirely. I’m not saying all obviously, but a lot of people hate TAA and some people hate it without even knowing what it is.
I’m advocating here for a lot of people. Do not develop Lumen with TAA in mind.
Lumen’s flicking problem is only going to perpetuate forced TAA and TAA dependence.
For instance, Death Stranding and all of its effects looks fantastic with or without temporal AA methods.
While more and more UE games are forcing TAA on to fix several effects, like temporal hair, AO and water effect(TAA dependance, turn it off: The game look even worse with vibrating or flickering pixels).
I’m not saying you are not right, but maybe there’s something lost in translation. Maybe I’m not looking at the right spots. But one of the images has a bad case of aliasing, and the other looks all right.
Images posted with 200% zoom and also a gif animation with the comparison.
1 Like
@TheKJ , why speaking about TAA now, when it’s been the default antialiasing method (and not only in UE!) for many years? Even in UE4… for me, it’s my preferred AA, combined with a slightly sharpness post process. Never seen any artifacts, nor blur (you can disable the motion blur effect, if you want, and the image is quite clear for me). I think someone must be very special to notice any artifact in a normal usage… and I’m absolutely hard when searching for visual incoherencies.
But today you have even “better” alternatives like TSR, DLAA, AMD one… but obviously you won’t ever have all the things at the same time. There is always a tradeoff.
Anyway guys: TAA is not related to Lumen, and this is the Lumen thread. It’s already quite difficult to be read by the team here, so let’s keep this as clean as we can, if you agree, with direct feedback about just Lumen, please.
Regards!
I think we as a community haven’t done the best job keeping the thread clean and focused on lumen specifically, as opposed to various light transport conversations/technology comparisons. I’m sorry for my role in that as it is. Don’t want to make your job any harder than it already is.
I tested it myself at 1080p, and it seemed quite usable all things considered. Not perfect, but the game was legible and I had no issues with the scene.
Now that I would like to learn more about, because as far as I understand it, nanite almost always performs better with diffuse, opaque geo. Performance is non-trivial with WPO and anything similar, but that’s not a new issue. Besides, lumen effectively depends on geo for the surface cache to run as well as it is, as it’s constantly generating and regenerating mesh cards via nanite.
In news directly related to lumen and its’ featuresets:
Fresh off the light transport presses as of yesterday, it’ll be interesting to see if/to what extent a ReSTIR-based integrator will offer for improvements. Obviously a prototype, but I’m very excited nonetheless.
4 Likes
Hi Snake,
I still don’t see TAA related to Lumen. Lumen noise, ghosting, and rest of problems are Lumen issues, so they are still there even if you have TAA, TSR, or AA disabled. But, I’m curious, what’s your proposal? As, as you may read, other AA methods are not working well with deferred rendering. So you have the option to use TAA, DLAA, etc, or nothing. Bring that options to your users.
Anyway, sure, I also don’t like Lumen to be so temporal-accumulation dependent (not talking about any AntiAliasing method). So, Lumen needs many frames to be clean, so you need a super high framerate, however, Lumen is capable to achieve low framerates… it’s like a paradox.
I also don’t think about games, in general, being released with Lumen nowadays. Just a personal conclusion: I think you should take Lumen and the current state of UE5 as a developer state (as not only Lumen, but it’s quite broken in many areas are even elemental features, like moving assets inside the Content Browser). Any game should be released with it in the next 1-2 years, probably, but you can already start developing it. Let’s wait. It’s a growing development and it needs time. I can’t wait too, but there is no other way.
As a suggestion for you forum activity (always from my point of view and without trying to hust anybody): for me, at least, it’s very hard to dig into your comments; they are so dense, with tons of collapsed texts (aka ‘spoilers’), tons of external links and tons of text, that we need like a half a day to read just only one of the posts (not even following each link, which link to other links, etc). I would suggest you to be more direct, just with few phrases, posting the images inside your post, not linking to 15 other posts, and being less repetitive. If preferred, you could also make an unique thread/post with all your detailed and dense thoughts and just refer always to that post, instead of writing lot of similar posts and linking one to each other, infinitely. Better to say something like: “Ey guys, I would want to insist in the temporal accumulation issue, here is the link to my thoughts-thread, which I have edited with additional info”.
Best regards
2 Likes
Hi @jblackwell is that in the latest main branch ?
1 Like
I basically use Restir GI and the legacy RT reflection(including Reflection Restir GI) as well as Mesh Caustics in the City Sample: City_NvRTX_Causitcs_521_Overdrive.rar - Google Drive
Hello everyone! First time posting here, sorry in advance if I’m making any mistakes. 
Since UE5.2, I’ve noticed a disparity between scalabilities, especially noticeable indoors. Currently, in “High” settings, there is excessive Exponential Height Fog leakage compared to “Epic,” which looks like it did in UE5.1 in both scalabilities. I’ve been going through the console variables that change between scalabilities and found that r.Lumen.TranslucencyVolume.TraceFromVolume 1 is the one that eliminates this effect.
The real question is that in UE5.1, the same cvar is configured the same way as in 5.2 but doesn’t exhibit this fog leakage in “High”. So, I have a feeling that I might be missing something. Any comments are welcome, I’ll leave the images down below for a clearer understanding.
Thank you very much, and I hope you all have a great day!
1 Like
It is indeed!
I tried opening the.rar and it refused to open for some reason, it tried decompressing and then quit on me. I’ll need to figure out why, but I’m just letting you know.
Unreal Engine 5 - RTX OFF vs RTX ON!
both are same! 
What are you even testing or benchmarking? RTXGI vs what? Baked lighting? Why is it in the Lumen GI thread? Why DX11 vs DX12 when you’re testing RTXGI on vs off?
1 Like
Lumen team, do not waste time reading my insufficient(to your work) reply to Farshid.
It’s just baked raytracing? The left isn’t computing anything anymore hence the massive perf gain.
If you move an object from the left side, shadows and bounced light will stay static.
In the right one, shadows, AO, bounce light will update in real time, which is ideal for games.