UE 5.4.0 is much slower than UE 5.3.2 on Mac Metal

I am trying UE 5.4 Preview, create a TopDown game from template without any modidication. then click the play button to run simulation. Only 8 frames per second, and it responds to my mouse events with lag.

But if I try the same template in UE 5.3.2, it is much fast. ~25 frames per second.

Mac OS M1 / 16G RAM

I see the MTLPP in v5.3.2 is replaced by MetalCPP in v5.4. Could that be the reason? Is there some project setting to solve it?

3 Likes

16 GB sounds like not very much memory for Unreal Editor TBH.
That being said, maybe the Preview is built in debug mode?
Have you run the profiler to see where it’s spending the time?

2 Likes

I don’t think 16GB RAM is the bottleneck here because v5.3.2 works well. And it is just a very simple game created from template. I don’t add anything.

The bottleneck seems to be from GPU, I tried to trace but don’t have a good understand about what the graph means, seems render frames take much longer.
When the game runs, GPU hit 100% while CPU is only ~10%; RAM usage is green.

I am using the engine downloaded from launcher.
In general, operations are the same, I tried version 5.3.2 and 5.4 Preview

  • Create a simple game from template without any modification.
  • The v5.4 Preview is much slower comparing to v5.3.2.
  • I tried packaging the game, same.

From what I see from source code, v5.4 dropped MTLPP and replaced it with MetalCPP, could that be related?

Does anyone else observe same slowness in v5.4 on Mac Metal?

And I see this warning messages occationally

LogMetal: Display: Timed out while waiting for GPU to catch up. (500 ms)

Tried in another Mac M2 machine, same. v5.4 is slower than v5.3.2

1 Like

When i make render it takes too long each frame to render. On UE 5.3.2 each frame would take about 2 to 4 seconds to render, now with UE 5.4 it takes 1 to 2 minutes to render each frame. Way too slow!

1 Like

Tried in latest official 5.4.0 Release, same. I can even feel the editor slate UI is slow to response

2 Likes

Not only it is much slower, but nanite which was working(although not performing very well) in previous release is completely broken now. I cannot believe this is an official release. Why even release it in this state, I’m 100% nobody tested this before releasing.

I was so hoping for new features and better performance on m3 and instead they made it slower and broke what was working, very disappointing and nobody cares to comment on why is it like that, or when it will be fixed or even recognising the issues.

1 Like

Mac studio m2 max here 32gb , this version of engine is 50% slower on my system. Also local fog volume is simply gone from my project, and when putting new one its gliched out. Not working

1 Like

seems like its becoming a standard in software industry

1 Like

The devs suggested this may be a TSR issue—could you try setting r.AntiAliasingMethod=2 and let us know if that helps?

1 Like

So I don’t know about guys here using M1 or M2 but in my case with M3 changing to TAA or even disabling AA using r.AntialiasingMethod doesn’t do anything for performance. Even completely new clean project has same performance with or without TSR it looks like…

Same goes for multiple older projects where clearly the perf. is drastically worse… I always use nanite though and since that is clearly broken and glitching on m3 and sonoma that might also contribute to some poor performance.

Before UE5.4, I could initiate a render scene in the project and seamlessly switch to other software, such as DaVinci Resolve, without any issues. The workflow was incredibly efficient. However, since the update, working while UE5.4 is rendering has become impossible. I’m unsure of the exact reason behind this change, but it’s disappointing that UE5.4 has become completely unusable in its current state. The render times have dramatically increased from 2 to 4 seconds per frame to a staggering 1 to 2 minutes per frame. This significant slowdown is severely impacting the workflow and productivity, making it difficult to achieve timely results.

1 Like

Mac studio m2 max and same here, changes nothing at least its not noticeable to me. I am making very simple project with baked lighting in door small area and performance dropped by 50% in 5.4. and new project templates has the same issue. even blank ones.

Edit: By adjusting graphical settings I have managed to reach acceptable frame rates. not sure if the graphics were updated hard to see what was updated without someone pointing it out. Maybe not the right project.

1 Like

I had problem not mac but it chose my onboard card to run might be worth a look this fix all my problems.

I have a large terrain with heavy sculpting with very high areas, lots of meshes with some heavy materials, auto mat on terrain, and get 74fps average if I don’t go ‘auto’ . Its quite doable in 5.3, 5.4 same project, fps tanks to 24fps. ryzen 3 3600 rx 6600 8gb 16gb ram very fast m.2 hd.

Same here, ur 5.4 wit5h ue 5.3 project and 50% as you noted. Just horrible, and I shouldn’t have top fix it with graphics settings. I have a semi massive terrain and lots of objects, some heavy materials, and 5.3 towers over 5.4. btw, I wonder what happened to better rendering, giving us multi-cpu hardware making rendering 50% or so faster…hmmm

So, not a Mac, then?

Also, 16 GB system RAM absolutely isn’t enough, especially with the backing store for the GPU taking half of that. The rule of thumb is to have at least 4x system RAM compared to VRAM.

Same here. It is unusable. Needs an urgent update.

Hey guys, I’m running Unreal 5.4.1 WINDOWS from source today, I improved editor performance by disabling Studio Telemetry, UDP and TCP messaging plugins, however, I still note that in editor I have lost 20fps from a 60fps scene and that my packaged performance is terrible. I have my game heavily optimised it has been from the start, I’ve been diligent about performance, measuring and finely balancing thread impacts as I go. Another example: where I was getting 70-90 fps in my open world with 5.3.2, now I get only 30~ fps. The thing to note here as well is that this is the frame rate when looking at the entire scene and when looking at the sky. As a test whenever looking at the sky I’ve always got 120 fps in previous engine versions.

If you enable the performance and profiling overlay, where does it say the time is spent?
Does it depend on size of window/viewport or not?

Marginal gains based on viewscreen size, I use a quarter viewscreen on a 2560x1440 display area, however all my performance marks come from packaging. My workstream is to always package after any change and test performance, hence why I had already managed to take UE5 to a massive open world with a great framerate. Across the board everything is slower now at runtime and despite massive optimisations that now effect quality, I am unable to get as good a framerate as in 5.3. Funnily enough I packaged my game back in 5.4 prior to preview, the frame rate way back then was the exact same, I therefore concluded at that time, the performance enhancements were yet to be committed to the main branch. Another example: In my open world render velocities has gone absolutely mad, now at 9-11m/s in my scene. I’ve never even seen render velocities in the profiler before that’s how small they used to be. I’ve been trawling the internet to try and work out what might be the cause of render velocities and as far as I can tell that is motion blur and temporal AA and Screen Space Reflections, however if I turn off anti-aliasing, reduce motion blur amount to 0, toggle motion blur off and then switch to no reflections I get exactly the same render velocities as I did running full lumen for lightning, Screen space for reflections, or indeed Lumen Lightning and Lumen reflections… Example 2: In my single landscape level where I have just a deathmatch level consisting of buildings, when I first loaded unreal 5.4 with the same config settings I’ve had in 5.3 + 5.2 I was at 20fps in the editor. After making optimisations to dynamic shadows and optimising reflections in project settings in editor I can now see 120 fps in the viewscreen there, however though my runtime performance is now more stable in that map, it is lower, despite the massive optimisations. Installing old tool chain today, I’m going back to 5.3 where I will be able to produce some real side by side tests hopefully later this afternoon!