From Godot Streaming to UE5 Templates

Hi everyone,
I wanted to connect a few dots from some recent conversations I’ve had here.

I recently posted about asset streaming logic, and in another thread, I was looking into upgrading my PC.

An experienced members gave me some great advice about the current hardware market—essentially, unless you absolutely need it for specific high-end production, you either have to be incredibly rich or just comfortable wasting money to build a new rig right now.

Taking that advice to heart, I decided to stick with my current Mac, save my money, and focus on the indie path.

To close the loop on my curiosity about asset streaming architectures, I decided to fire up a few versions of UE5—specifically from 5.2 onward, as those versions introduced native support for Apple Silicon.
I ran the native macOS builds across different versions of the default Third Person Template to see how the frame rates hold up at default settings.
Below are screenshots of the GPU statistics showing how each version performed.

In the top-right corner, you’ll see the standard built-in Apple Metal HUD.
(Note: UE 5.2.1 seems to have broken built-in statistics, and 5.5.4 didn’t load them at all.)

Newer shaders and heavier features like dynamic resolution, Nanite, and expanded virtual texturing are significantly deteriorating performance on this hardware as the engine evolves.

SM5 for 5.2.1, 5.4.4, 5.5.4
SM6 for 5.7, 5.8

With yesterday’s release of Godot 4.7 bringing more vibrant lighting and a ton of other advanced features, have you noticed a similar trend of adding more heavy components by default? Is it taking away from Godot’s signature lightness?

I would love to do a similar, structured test for Godot, but I want to make it fair. What do you think should be included in a Godot template to accurately reflect similar rendering conditions to UE5’s Third Person Template?

Size of projects

Packaged executables

No, nothing was added in the 4.7 that by default is more demanding than in versions before. If something is too “expensive” (calculation wise) than don’t use it in your project (like VoxelGI, AreaLight3D, SSR, SSI a.s.o.).. You always have to decide beforehand what your targeted hardware will look like; test, test, test on that hardware. And offer settings / options to the user to setup the rendering. It’s the same in other engines. Someone with a RTX5090 can crank up the settings and rendering features more, than someone with a 1080, or on other hardware.
The “UE5’s Third Person Template” is just a template, with settings that will fit average hardware, it’s one of many possible looks.
IMO, there is no set of features that will fit all visual styles. Choose a more stylized look for your projects and fake (baking etc.) where it’s possible.
Maybe target more at a benchmark where you can measure the different render settings / features and rank them how demanding they are on your hardware.

1 Like

Your point about focusing on a feature-by-feature benchmark rather than a “template vs. template” comparison is a smart pivot. Since my goal is to maximize efficiency on Apple Silicon without spending a fortune on a new PC, isolating the exact cost of individual settings (like SDFGI vs. baked lighting) seems like a much better approach.

Thanks for the perspective—it definitely changes how I’ll structure my testing!

Good point on the visual style, too. I’m actually planning to use free low-poly assets from Kenney and KayKit. They have a great, recognizable look and use tiny texture atlases, making them incredibly lightweight out of the box.

Using those will make it much easier to isolate and benchmark purely how the engine’s rendering settings (like lighting and shadows) impact performance on Apple Silicon, without heavy geometry or textures muddying the data.

1 Like