itch.io

Follow-up: I've now exported the demo to web and measured it in a browser.

- Setup: Godot 4.7.2, Web export with the no-threads release template (so no COOP/COEP headers needed), Compatibility renderer (WebGL 2). Chromium-based browser on Windows, same i5-13450HX + RTX 4050 laptop.

- It runs. No errors or shader warnings in the console; flash and dissolve compile and apply under WebGL 2.

- CPU time per frame (effects looping non-stop, each frame stepping 1/60 s like a 60 fps game; desktop Compatibility in brackets, same method): 1 large explosion ~3 ms [1-2]; 5 ~6 ms [2.5-4]; 10 ~11 ms [4-6.5]; 30 ~23-30 ms [11-16]; 30 small impacts ~4-6 ms [2-3.5]. GDScript in WebAssembly came out about 2x the desktop cost on this machine.

- What that means: in a browser, up to roughly 10 large explosions on screen at once fit in a 60 fps frame, with not much room left for the rest of your game. 30 large ones at once don't (that's around 35-40 fps). Small impacts are cheap. Slower machines will be worse.

- Correction to my earlier numbers: the desktop figures I gave before (e.g. 30 explosions ~4-4.5 ms) came from a benchmark running at several hundred fps, which averages toward the cheap frames. At a fixed 60 fps step the same desktop is 11-16 ms for 30 explosions. Sorry about that; the numbers above use the corrected method.

- Not tested: Firefox, Safari, mobile browsers, integrated GPUs, and the threaded web template.

The free sample uses the same drawing code, so a quick web export of it in your project is the most reliable check.

Good question. What I measured and what I haven't:

- Compatibility renderer: yes, on desktop. The demo project actually ships set to Compatibility. I ran the full demo and every effect under Compatibility (OpenGL 3.3), Mobile and Forward+ on Godot 4.7.2: both shaders (flash, dissolve) compile, no errors or warnings in the log.

- Cost (Compatibility, desktop, whole-frame time, i5-13450HX + RTX 4050 laptop, effects looping non-stop): 1 explosion ~1.2 ms, 5 ~1.8 ms, 10 ~2-4 ms, 30 ~4-4.5 ms; 30 small impacts ~2 ms. Forward+/Mobile were 0.3-1.3 ms faster in the heavy cases; most of the cost is the GDScript painting, which is the same in every renderer. An earlier run of the same benchmark gave higher numbers (10 explosions 6-9 ms, 30 about 17 ms), so treat these as a range, not a promise.

- Web export: not tested yet. I haven't exported it or run it in a browser, so I have no web numbers. The code uses no threads, no file access and no OS calls, and the shaders are plain canvas_item shaders, so I don't expect a blocker, but GDScript running in WebAssembly will be slower than the desktop numbers above, and I can't say by how much until I measure it.

- One known detail: the dissolve noise pattern differs slightly between GPUs (it uses a sin-based hash). It still works, it just won't be pixel-identical across GPUs.

The free sample (impact, sparks, ring, flash) is the same code as the full pack, so you can try it in your own web build before buying.