UE 5.2 vs UE 5.3 big performance loss

Upgrade from UE 5.2 to 5.3 resulted in
+25-30ms VolumetricFog rendering time
+7-10ms LumenReflection rendering time

Anybody else with this issues? I found no visual upgrade in my scene and found the performance impact on both AMD and Nvidia. Lumen got some changes in 5.3 in the release notes regarding reflection but no setting to get the 5.2 performance back. I’ve no clue why the fog got incredible worse. I know that volumetric fog costs a good amount of performance (so does Lumen) but it’s fast in 5.2 and very bad in 5.3

4 Likes

Yes, the same problem, with TSR from 60fps bacame 30fps, and on MSAA from 120 to 60. But MSAA looks much much better now :slight_smile:

Nvidia GeForce 3060 Ti, 4k resolution

1 Like

To post something with a bit more values.

UE5 ThirdPerson template map. Volumetric turned on at ExponentialHeightFog. Everything else is unmodified.

UE5.2.1: 79fps
UE5.3.0: 49fps → no visual improvement at all

GPU: AMD RX 6700 XT (but also tested with Nvidia card… same impact)


UE5.3 makes volumetric fog unusable. This + changes with Lumen ruins the performance. In my opinion 5.3 is not production ready.

6 Likes

Hey, I got the same issue and found the cause for me at least.
In the previous versions of the engine, it would by default upscale post processing and lighting, where as now it instead renders them at (at least) full resolution. It appears a lot of scaling functionality has been redone and despite inheriting the old settings, those settings don’t affect things like PIE the same way any more.
Note that some engine scalability settings may also have changed, E.G as of 5.2 AA at Epic is 200% render scale rather than 100%.

5.1 vs 5.3:

Sadly, I have yet to find a specific setting to replicate the old behaviour aside from just setting the viewport scale, but that affects render scale/resolution as a whole and not just GI and PP, and is separate from the packaged game.

(Frankly, there’s too many settings you can set from too many different places that affect the various systems in different conflicting ways now, and what group of settings is intended for which situation is very unclear and a mess, viewport settings need to be clearly seperated from game settings, then you have editor vs engine vs project and console commands on top of that, as well as additional duplicates depending on what menu you access.)

It’s a shame because while it’s technically higher fidelity, there’s really no noticeable difference at least with my game’s artstyle, but a much, much worse framerate.

5 Likes

I have just been sticking to 5.2 but has anyone figured out why 5.3 is like 20 fps slower than 5.2?

2 Likes

I checked again at 5.3.2 and compared with the same project in 5.2.1.
In a nutshell: I think 5.3.x is similar to 5.2.1 (even if FPS are worse by default in the editor).

A bit more detailed:
5.3.2 default editor performance is still a lot worse to 5.2.1 default editor performance. But 5.2.1 is fake. While I compared settings from 5.3.2 with 5.2.1 I tried changing the ResolutionScale (r.ScreenPercentage). While I immediately saw a difference in 5.3.2 (starting from bad FPS and getting better and better) there was no difference at all while moving the slider in 5.2.1 (no matter if I set it to 100% or 30%).

The reson for that was spammed in the console of 5.2.1: “r.Editor.Viewport.OverridePIEScreenPercentage” is set to 1 so it always ignored my resolutionscale changes and kept it to some strange “editor-100%”. Switching this to r.Editor.Viewport.OverridePIEScreenPercentage to 0 in 5.2.1 gives me pretty much the same (bad) FPS as I get in 5.3.2 with 100% resolution scale as well. So for me it seems the 100% ScreenPercentage in 5.2.1 with disabled override differ from the 100% ScreenPercentage with enabled override. Probably some other settings are kept in this mode as well so I’m not 100% sure if something else changes. But I would not run the game finally in the editor anyway so only the non-overriden stats matter.

5 Likes

I am seeing something similar in 5.3.2 from 5.2 the FPS is 50 frames different after importing the same landscape file between the 2. In 5.2 I get 120 FPS with an 8K terrain. In 5.3 I get 70 FPS. There is something different in the versions for sure.

2 Likes

Yes, @Neutronux nailed it on this one. EDIT: I’ve decided to simplify my original answer to avoid causing confusion. 5.3.2 may SEEM to produce worse performance out of the box, but that is only because 5.2.1 was NOT respecting your Screen Percentage settings and was always downscaling the output.

TL;DR: 5.3.2 has excellent performance once you match the Screen Percentage settings that earlier versions were using by default. In fact, 5.3.2 has even BETTER performance than 4.27.2 at true 100% Screen Percentage. I’m currently using 5.3.2 and like it. :smiley:

On older GPUs, you can use these Project settings to get the best performance:

  • Dynamic Global Illumination Method: (None)
  • Reflection Method: (Screen Space)
  • Shadow Map Method: (Shadow Maps)
  • Nanite: (Disabled) (also added r.Nanite=0 under [/Script/Engine.RendererSettings] in project’s DefaultEngine.ini)
  • Anti-Aliasing Method: (Temporal Anti-Aliasing (TAA))
  • Temporal Upsampling: (Enabled)
  • Early Z-pass: (Opaque and masked meshes)
  • Mask material only in early Z-pass: (Enabled)
    (Those last 2 settings help with dense foliage)
  • Allow High DPI in Game Mode (Enabled)
  • Default screen percentage mode for realtime editor viewports: (Based on display resolution)
  • Default RHI: (DirectX 11)

Some additional notes showing differences between UE 5.2.1 and UE 5.3.2:

  • UE 5.2.1 has much darker shadow rendering than UE 5.3.2 using the same test settings. With a few tweaks, I don’t mind the look of UE 5.3.2, and I prefer the warmer sunlight in UE5 over the way UE4 renders the same scene.

  • Apparently, in UE 5.3.2, Epic changed the Camera’s default Field of View to maintain Y FOV instead of X FOV. In order to match the look of 5.2.1 (and earlier versions), inside your character’s BP, go to Camera Options, check “Override Aspect Ratio Axis Constraint” and leave “Aspect Ratio Axis Constraint” as “Maintain X-Axis FOV” as shown in this answer here.

  • In UE 5.3.2, You can no longer change the Screen Percentage setting before launching a Play In Editor session, which is disappointing. However, once in a Play session, there are new handy presets you can select to quickly increase or decrease your Screen Percentage.

  • To improve your viewport framerate (without launching a Play session), you can click the 3 little lines to the left of the current Camera setting, choose “Screen Percentage”, select “Custom Override”, and choose a lower value according to your preference.

5 Likes

so this means 5.2 is way more better than 5.3 in terms of fps am i right?

Just use 5.3. Yes, you would get worse editor perfomance. But this does not matter that much. You have to check you FPS if you start a standalone game without editor. The editor in 5.2 is overriding your resolution settings.

Is there any way to improve the performance of 5.3 to 5.2 levels? I have a fast 16-core 4.4ghz PC and an RTX 3090. The frame rate in the editor is low.

Try GetGameUserSettings → SetResolutionScaleValue … and choose a lower value. It’s pretty much what your human players would have to do when it’s finally built anyway (most likely via some in game menu). Also if you use TSR you should lower the settings (GetGameUserSettings → SetAntiAliasingQuality). I’m not 100% sure which additional settings are overridden by default in 5.2 (there might be more) but this should bring you close to the (worse) quality and the (better) performance of 5.2 editor again.

1 Like

Yes, but that lowers the quality/resolution of everything. Is it not possible to downscale individual render passes like they seem to have done in 5.2. For instance a light pass at 50% etc?

Anyone has compared the shipping builds of 5.2.1 vs 5.3.2 with Lumen and Nanite off?
Upgraded a copy of my project from 5.2.1 to 5.3.2, fps dropped to 60% then uninstalled the 5.3.2 version before testing the shipping build
Comparing the shipping builds is needed if the performance difference is supposed to be the default setting of screen percentage mode for editor viewports

I think my last post may have introduced confusion regarding performance, so I have edited and shortened the post to clarify.

This topic was automatically closed 30 days after the last reply. New replies are no longer allowed.