Unreal Engine 5.3 Released

Anyone tried the Hair Strands continuous LOD?

Before 5.3 many slate properties got getter / setters and direct access was deprecated. For my sanity, I am not going to check right now if they added the missing getter / setters from 5.2. I reported that stuff here:

[Bugs bugs bugs] Slate widget synchronization - #6 by Roy_Wierer.Seda145

How do you fix this?

Anyone else having glitches with animations having additive curves modifying the root bone?

New nodes replace the old nodes, not sure why they just didn’t convert the old nodes.

Even reporting them, they will be (usually) ignored. I have been reporting issues since years, tons of them. Even something as elemental as moving/reordering assets inside the content brower is broken. I reported it months ago, within other bugs too. Still no reply nor bug tracking number. Even I thought I could be in a blacklist, as a ‘spammer’ or something (I’m doing a free reporting work!), but also tried from another account with the same result…

Sending a bug report requires testing, simplification of the steps, and a time to clean, redact and send. Too much for getting ignored. From my side, I’m not going to be a free betatester and do it anymore for a while. Good work Epic.

At least, writing here, people can complain about it in a fast and easy way, and inform/warn other users about how much Unreal 5 is broken. Too much hype and marketing about it, but please, use UE4 instead, as I already recommends to all my big (industry leaders) clients, after encountering many many walls in every simple development (which didn’t was in UE4, of course, even if some things were a little broken too, but 5 it’s like a joke). They all are disappointed with UE5 and I didn’t need to convince them about it.

Please, Epic, what the hell are you doing/ what’s happening? An official and explanatory reply would be so useful and inspiring. Much better if mentioning a possible stable future specific number release. I was thinking it were 5.3… but still waiting.

4 Likes

I think the getters weren’t there. My overridden class added the getters to protected members. Then they moved the DLL Export from the whole class to specific functions, which made the protected member vars no longer available outside the module. https://github.com/EpicGames/UnrealEngine/commit/f4c1a0307ca9900606d2741a4fb063681b309d57

At this point I might as well make a proper pull request to add this and see if it gets accepted. Unfortunately pull requests also go ignored for a long time so I’ve stopped putting in effort into changing the engine and instead add overrides of engine classes to my game. However that’s not possible if things I relied on being exposed or DLL exported are no longer available.

This seems to be a common issue too, like many things are private or not exposed that should totally be. It’s not hard to add a getter to expose the ScrollBar on a tile view so I can have more control over it, but they don’t for some reason. I’d prefer to have no engine changes in a private fork at all so it’s easier to always pull in the latest version of the engine for my game. But with this, it’s going to be at the point where I’m gonna have to have all these random fixes to the engine, and merging releases becomes a nightmare.

I used to work with UE4 since 4.0 and had all sorts of pull requests get merged in back in the day. It’s a bit frustrating that even simple one line fixes for huge bugs go ignored for years.

Here’s an interesting example:

These public methods don’t have a DLL export, so I’m hosed without modifying the engine source and making a Pull request adding a dll export. I sure hope that PR gets accepted.

2 Likes

I am in exactly the same situation. Same issues on UE4 though, 5 was not a rewrite of 4, just an addition of more experimental features.

1 Like

Yep. Agree, but a regression in many cases too

1 Like

Is this only available for Linux?
Will it be available for windows anytime soon?

I’m working on a big Unreal Engine 5.2.1 project, specifically on the Apple Silicon Port of the game. It would be good to know how is performance looking on 5.3 vs 5.2 considering nanite and world partition. Also is SM6 working on Mac already ?

Building UE 5.3 with disabled Unity builds is making a tons of build errors in Binary engine.
This can be reproduced by giving the -DisableUnity option when doing build from UBT, or by setting the <bUseUnityBuild>false</bUseUnityBuild> inside the BuildConfigurations.xml file

this looks great, so is the nanites work on splines and foliage?

Good, I’ll be testing the performance for improvement checking.

Can you guys make UE5.4 focused on game performance now?
NEW UE5.4+ feedback: Start working on and investing in PERFORMANCE innovations for actual GAMES.

I may have posted this before but It has been edited and updated with many more issues and problems with the UE5 mindset. Specifically Nanite and these fast links:

“Fast links to my post and replies related to this topic.
Post #2–Definition of optimizing.
Post #3–AI workflow for UE5 for optimizing static meshes. -Possibly the best investment Epic could make in.
Post #4–Why 30fps is not acceptable and where FN’s 6 billion revenue should go.”

I understand that Nanite is supposed to help developers not “waste time” developing LODS.
But Nanite is not the way. Take a wall mesh from the City Sample. It was something like 200,000k tris? Just because Nanite makes it usable doesn’t mean we should?
(Btw even if the mesh is lower poly, it will have worse performance with Nanite)

Using the modeling tool, I decided to scale down the mesh to around 4500K tris using the UE standard simplifier.

Spoiler alert the mesh looked awful. But if I had gone though the hour of work to optimize the mesh myself, I would have just flattened down some detail via baking in the detail with some fake parallax occlusions.

Mostly likely with some time and miner intelligence, I probably could have scaled down that modular building piece down into a very good looking 4000k- tris mesh. Yes some detail would be lost geometrically but still see-able in a gameplay situation.

Why go through the work? Because it will beat Nanite out of the park with performance.
Now a lot of studios are not going to hire a someone to do the monkey job(Legit, making a high poly actually game ready is a long tedious project that requires minor intelligence and training).

Minor intelligence and training but it takes a long time? Why the hell are we not handing this workflow off to AI!
The Nanite workflow is destructive to consumers due to the performance overhead of Nanite. It only servers developers. NOT GAMERS. And maybe virtual production.

The AI workflow in the post that can massively help developers the same way Nanite saves them time. And the performance boost from the intelligent meshes would help gamers.

This is one idea not far-fetched at this point. Other things tht could help performance is by making more optimization in Lumen and VSM’s in computing efficiency via more focus calculations on actor mobility.

Hi @locateli ,
The 5.3 release notes give the answer
5.3 nanite notes

Hello, I need help…
After a fresh install and unreal 5.3 not opening…i uninstalled epic games launcher and unreal, 5.3.0. Deleted everything off the Mac Studio, M2 Max. I proceeded to reinstall epic games launcher, and then unreal 5.3.0. Before launch I restarted the computer. I opened up, epic games launcher. I ran unreal engine 5.3.0, this time the unreal logo was jumping up and down for a minute, but then resolved to nothing. I tried again, now this time like the first the unreal logo pops up for a second and disappears, nothing happens.

Please someone help

Why 5.3 so slow when trying to make LODS and meshes editing. It’s barely using any of my 10600k.


It’s been like this for 7 mins. (EDIT 15+ mins now, getting ready to go end task with task manager)
5.2 did not act this unoptimized.

The… performance overhead of Nanite? Are you crazy? Automatic pixel perfect LOD will give a far more optimized scene. Not rendering the obscured parts of the 3D models too.

3 Likes

I DID the TEST.
And Epic Games keeps promoting Nanite like it’s to God’s gift to rendering meshes.

If FN and City Sample perform the same, that is a clear sign of a LACK of scene specific optimization.

Nanite should not be the way we “Optimize meshes”.
Hands down there is a smarter way.
(Working on the 5.3 LOD vs Nanite test)

EDIT:

A new test

With Nanite-10.4k triangles

And without Nanite, and very bad 5 LODs levels-3.5m tris.(That is a lot more than my other test). Both test where done with a 10,000 tris mesh.

What do the test results mean?

Yes, for high poly meshes scaled down to a reasonable size, Nanite will render faster.

But, if you were to optimize the mesh better, like baking in it’s detail, and creating hand made/more intelligent LODs(Not just crappy ones like I used in the test that half triangle count).
We would have much better performance than just slapping down Nanite.
And that is why Nanite is going to kick gamers in the face unless we innovate further about how we optimize meshes.

(In the original Nanite vs LOD test, I worked pretty hard on the LODs to make sure they stayed under 1m tris)

Like I’ve said before. Really feel like this is a job for AI. And I’m talking about this is for 100% static, environment meshes. Not moving objects. It’s almost the same workflow Nanite offers but benefits more people.

I see this as the answer so clearly to the problems developers face with time and money on LOD work but this also solves the problems gamers are having with such poor rasterization performance due to Nanites over-use/head.

The… performance overhead of Nanite? Are you crazy?

@Altrue Just noticed you are an engine contributor. Read my Nanite post. I am not some random single developer who knows this to be the truth. You are wrong about Nanite and you guys need to STOP revolving features around it.

I just realized this class is a template so dll exports here should be unnecessary, not sure why the functions aren’t found.