Godot GLES2 3D Benchmark on real phone Samsung

This is a READ-ONLY snapshot of the Godot Community Forums when TwistedTwigleg was the owner, which is hosted for archival reasons only. The current Godot Community Forums can be found here.

Some links may point to external websites, including the current Godot Community Forums. These websites are not associated with TwistedTwigleg.

Godot GLES2 3D Benchmark on real phone Samsung

CristianoSimao

When we read in the Godot documentation that mobile have a limited hardware, the question in mind is.. "How many triangles I can use in my game for mobile phone?"
In this video I show a test in a old Samsung Galaxy J5 and the results.

Link to You Tube:

In the reality, for mobile you need uses less triangles as possible.
And in this mobile phone the hardware is very similar to Playstation 1
Than research all details from the hardware of the phones, so you can have a base to determine the "maximum" graphics of your game.

Megalomaniak

If you're trying to just test how many triangles can be drawn then your test might be flawed since by adding lots of low-low-low-poly objects/nodes you might actually be bottlenecking the cpu instead of the gpu with drawcalls and scene/node management.

Pixophir

Yeah, it is not simple. I suppose you can render millions of triangles from a huge buffer, indexed, with a single draw call and flat shading even on a mobile. And potentially tens of millions with instancing. But when you start rendering immediate mode, or emitting many draw calls in a loop, switching pipelines and other states, multiple render passes, processing and exchanging data between GPU and CPU, things can go down quickly. You might even end up with just a few thousand triangles at reasonable frame rates. So the mere number of triangles without knowing what's exactly going on is not necessarily a good benchmark.

Books have been and are being written about what can be done performance wise.

cybereality

It's good as a baseline, and it's easy to test with various engines (just using spinning cubes) but it's not a realistic benchmark. Draw calls and shader/material complexity have more of an effect. For example, if you used instance geometry, you could probably combine 100,000 cubes and render them on a phone. At least I've done that on desktop with over 9,000 FPS (not a meme) so I'm sure my phone could do 90 FPS. And you have to consider draw calls (affected by material state changes). Like 6,000 cubes with the same material work, but you might only be able to do a few hundred cubes if they all had different materials. Also, the shaders involved, shadows, post processing, lighting, etc. Too many variables.