Unreal Engine 5.3 Released

General Announcements
,
89

First off, let me clarify that I’m very appreciative of the amount of work you put into this. But as I said in the other post, not counting Nanite’s occlusion detection performance gains isn’t fair nor realistic. As is not doing these tests in packaged games. But I understand this would be even more work.

Also, I want to amend my previous criticism in that there’s no performance overhead for Nanite. By definition to virtualize the geometry there will be a performance overhead. But it’s true that I do not expect the net performance gain to be negative in most (all?) realistic scenarios.

(Full disclaimer since you mention my badge, I did not contribute to UE’s rendering pipeline nor do I have the skills to)

90

not counting Nanite’s occlusion detection performance

Sadly it just isn’t enough! It’s clear from FN’s performance.

since you mention my badge, I did not contribute to UE’s rendering pipeline nor do I have the skills to

Fair enough, but my test and the other test from other people(included in my nanite thread) show LOD optimization will always blow Nanite out of the water performance wise.

And you know what? We need a performance boost?!
Epic Games’ focus on Nanite is sucking up the last bit of room we could have had for something that would actually help gamers not rely on blurry upscalers.
We need to help developers yes but not at the expense of kicking our customers/players in the face.
That needs to be the core of UE5’s next innovation.

91

Wow, well I realized one fix is to call

this->HasValidItemsSource() to fix the identifier not found error.

Wow, in fact, all the properties are accessible with “this - >” Phew, I’m glad I found a fix that’s actually stupid simple. I take back what I said about this being a huge issue.

Something about how I’m accessing from a template class confuses the compiler now so it needs an explicit this-> in my modules.

92

Check this out. I fixed it easily in my code after all.

93

What is the minimum download size for version 5.3 install?

94

hi @Only4Gamers 75 Gigabytes for Binary 240GB source builds 430GB on an SSD/needs SSD

95

I’m getting half FPS in linux (nobara) over 5.2 (nobara) and nanite does not work. I’ve tested 5.3 with wine and it works there, twice the fps, even with vulkan only enabled. Maybe it’s a driver issue, anyone has the same problem?

96

Managed to fix it: Disabled nanite in project settings then restarted the editor. Enabled nanite in project settings then restarted and it now works

97

Since the preview version, UE5.3 doesn’t work on my computer. Noticed that the same issue occurs with Fortnite since the update of the new season, which I believe uses UE5.3 as well.

In the case of Fortnite, the game crashes instantly if I enable Nanite (works normally if I disable Nanite).

But when it comes to the Unreal Engine editor, it simply doesn’t open. It opens the “loading screen” and then freezes, even when Nanite is disabled, and it shows as “Not Responding” in the task manager.

I’ve already tried disabling Nanite through configuration files, but it didn’t work. I don’t know what else to do, and I’m unable to test some products on the new version. : /

UE5.0, UE5.1, and UE5.2 work perfectly, but UE5.3 simply doesn’t open at all.

I will submit a bug report with my computer details and editor logs. : /

98

hi @lucoiso
The problems is probably due to the DX12 drivers that driver your graphics card the shaders probably cannot build and suspend
In 5.3 the default is DX12/ SM6.
You could try delete the shaders built onfirst startup
C:\Program Files
Epic Games\UE_5.3\Engine\DerivedDataCache\Compressed.ddp
I would follow the post which was earlier

#UE5 CRASH #UE5 Unreal Engine 5 Crash

You could try updating your Nvidia drivers need version 537.* is latest from GeForce Experience and would explain why Fornite has a problem

99

You should delete all ‘intermediate’ files such as the ones inside ‘intermediate’ folder, deriveddatacache - basically make it as new as possible. This of course doesn’t deny Epic has to do something here… but just to get things going…

100

The release of Unreal Engine 5.3 is exciting news for the world of game development and interactive media. Unreal Engine has been at the forefront of pushing the boundaries of what’s possible in creating stunning visuals and immersive experiences. With each new release, it’s amazing to see how Epic Games continues to enhance the capabilities and efficiency of the engine.

2 Likes
101

Sure… The release of Unreal Engine 5.3 is truly exciting news for the world of interactive media. I wish I could say the same for game development!

2 Likes
102

Has someone tested 5.3 performance vs 5.2 (? I just tested the Meerkat Demo(shipping build same settings) and I got these results:

5.2.1 / 5.3 (same frame comparison)

**51fps / 81fps +59% **
38fps / 53fps +39%

As far as I know 5.3 has optimizations to nanite, cloud rendering, chaos queries, a few with lumen and on top of this 5.3 is now using c++ 20. But it didnt expect such a massive difference, can someone please confirm my results ? (please don’t use a example project)

103

I have opposite (like from 60 fps on HD res without upscaling to 40fps (tested both: hithit lightning with fallbacks or SDF)). Using Nanite + VSM + VST + Lumen without TSR. But i tested in the game in PIE or in the editor. I didn’t try the shipping build, because it’s not worth putting energy into it and you can’t profile it on the fly. You will waste maximum time when there is something wrong, but it doesn’t even have to be.

“don’t use an example project”
“tested the Meerkat Demo”
You’re contradicting yourself a bit on this one :stuck_out_tongue:

I tested new project form scratch (half of them is low poly from UE4 waiting for new art,m half is done) + few scene made from assets from marketplace (eg: Pharaoh’s Legacy: Egyptian Temple Megapack, Oil rig, etc). But I had to redo them (adjust materials, adjust normals, wpo, redo lights (dynamics, complexity), throw out translucent faces, fix overdraw, some Nanite pieces for better cluster lodding and metrics, etc). However, things were instantiated in the level, not in packed level actors (object per instance). But I didn’t have one thing slower. Everything was slower (on gpu) + i updated drivers. But I’ll try later (proppably in UE5.4 or 5.5). Before that is definitely a waste of time for me.

I also tested it with Fluid Ninja live, it didn’t have much of an impact. Still works good.

ryzen 9 5950X 16-Core, 64gb ddr5, M.2 NVMe SSD, rtx 3060 (12 gb ddr6)

I haven’t tried the Merkaw demo. It is already quite old and I think static or stationary light with different albedo values, etc… or something like that is set by default (I don’t remember anymore).

From my experience, Nanite is the heel of the tech and good revolution in game art (but i still have problems with wpo and cluster culling (for example, I use custom calculated bounds per actor that I feed via blueprint to CPM into material for wpo or custom gradients with ISM components), or export from blender to fbx is worse than 3ds max (sometimes I break the mapping), sometimes I forget a vertex far from UVW (same result) + on 5.2 it is not possible to turn off lightmap generation per model!!). But I don’t mind. It still looks awesome, even if it’s not perfect (mesh pool is perfectly sufficientif if you don’t go crazy).

But VSM and Lumen are a bit of hell. Lumen looks good, but it take lots of performance, VSM has a lots of problems - eg: it is not translucent (if you don’t have higpoly, it creates really weird artifacts (small sharp triangle shadows or noise)).

When I needed to optimize it quickly, this helped the most:
image
I then overrride it in the game via the level script until I fix the menu.

So I stayed on 5.2 - migration is too risky.

104

ps: when someone finally does that f****g warning per model in map check, where if its use Nanite and not VSM, lots of developers will probably be grateful and not so confused by thousands and thousands of warnings without log filters. Based on the stream from Epic, I understood that they simply “left it there” and in my opinion that is irresponsible. People then does not notice other warnings, which are more relative to the project he is working on and can be much more serious.

105

Are you sure those numbers are correct? I tested the City Sample in both 5.2 and 5.3 and found no difference on the GPU side(Ai was turned off in both test and was not in editor.)

106

Yes, I’m absolutely sure about those numbers. I tested with Screen Space GI and Lumen. I changed some rendering options and in all scenarios 5.3 was absolutely beating 5.2.1 by no less than a 20% margin (including PIE).

You can try for yourself. One thing is that maybe the Meerkat Demo was changed somehow in 5.3 although I tried to find some discrepancy on the project settings and visual presentation, but I found none.

I’ll test another project later today.

107

Electric Dreams or City Sample could be good comparisons too. I hope to test them if I have some free time.

108

We also want better animation support and not getting the bending pipes

1 Like