OTOY Forums

Hi again.

I was wondering how memory works on a GPU. Does having more GPU memory increase the speed of the render, or does it just allow larger models to be worked on? In other words, does a GPU assign buffers only according to poly count, texture complexity and frame size, or can the GPU assign extra buffers to do extra work in parallel if it has the memory available?

treddie wrote:Does having more GPU memory increase the speed of the render

No. More memory = more storage. The speed of memory would have something to do with the loading speed at render start, but compared to the average render time, this effect is negligible.

btw, couldnt you post all this memory-related questions in one thread instead of having 4 or 5?

Heheh. Yep. I'm not thinking when I'm thinking.

Concerning large models (actually, models of any size, but large is where this seems important), does turning opacity down to zero effectively hide a surface from being evaluated, or does Octane still see it and process it even though it is invisible?

The surface is still there, and rays will still hit it.

One clear indication is in the kernel settings. The "alphashadows" setting means the shadow rays will take the opacity into account, so they will actually pass through transparent materials. Otherwise transparent materials still throw a shadow.

--
Roeland

The surface is still there, and rays will still hit it.

Then when multiple meshes are allowed later on, will there be the option to hide meshes when that happens?

Actually, come to think of it, that could easily be accomplished by simply disconnecting the desired mesh node from the rendertarget.

Hiding a lot of objects by using a material with transparency or a portal, will slow down your render time, sometimes considerably. So, you a re better off re-exporting the scene without those objects.

Continuing this thread on memory, can a GPU have its memory increased like on a motherboard or is it fixed and soldered in?

It's fixed, it's not possible to extend it.

--
Roeland

I thought that might be the case. Too bad since I think it would be a nifty idea for people who don't want to upgrade but just need an extra gigabyte or two of memory.