Dynamically loading Vrmesh in Vray RT GPU (CUDA)

[MuratCetin]

Hi,

It would be great if RT GPU (CUDA) could dynamically load .vrmesh data as soon as a ray (GI/shadow/reflection etc.) request surface info (RT CPU /Vray Adv can already able to do that).

I'm aware that RT CPU is very different from RT GPU, and also reading scene data from the disk on the fly can create narrowest bottleneck in GPU rendering history! But in current GPU implementation all vray meshes linked and loaded during the initialization. Although constantly loading/unloading geometry all the time might leave GPU sitting idle for too long, maybe it can be implemented in one direction i.e. load? So GPU does not have to load objects that are not visible at all. (by visible I mean "raytraced visible", multiple bounces, reflection refraction etc.)

Hey! you guys pulled off on the fly mipmapping for textures and it is incredibly useful. Why not show some love for geometry as well?

Thank you.

[savage309]

It is definetely doable for geometry, we just haven't got to that. We did it initially for textures, since the potential benefits there are bigger (we not only load them on demand, but we create mip-map levels for them runtime).

Best,
Blago.

[MuratCetin]

Yeah, it looks like a lot of optimizations going on in the background aside from mipmapping: I noticed that you trim channels for textures with full on alphas, drop grayscale textures to 1 channel. I can only imagine what you can come up with geometry optimizations .

Best Regards,
Murat.

[savage309]

Yes, we do that in 3.5 We will see in how it goes for geometry in the future.

Best,
Blago.