Yannick

Hey 3d folks, what is best for the GPU: having one shader updated multiple time per frames or have many instance which are only updated with camera transform (or other variable parameters) ? Cc @skaven_ @tomspilman @MonoGameTeam

I think it's more down to the vertex streams at that point. I think the shader gets cached and used as and when. I think this is why GPU instancing is so useful as it reduces the vertex stream massively...

Ok but is it a good idea to cache shader parameters? Because I can have a shader instance per material with 3 parameters changed per frame instead of all parameters. Don't know how @MonoGameTeam handle that internally.

There's no way to share cached uniforms across shaders, if that's what you're asking.

could you use structured buffers to pass params across shaders, like with compute?

No, MG does not expose it like that. Things like that have been discussed a long time ago, but we can't support it with the current shader pipeline.

Ok so using a cloned shader with parameters set one time is the same than the same shader used by multiple materials where parameters are updated very often?

Ah, I misunderstood. They're not the same. I think using 1 shader is faster. I would be interested to know!