System RAM - can in theory be used?

[voon]

This is just a technical question, not about current Octane. but in theory: Can a GPU rendere make use of System RAM to calculate bigger scenes? Some sort of swapping mechanism between VRAM and RAM? Or is it mandatory by default, that the whole scene is always entirely in VRAM, or the algorithms simply won't work/can't work etc?

[ROUBAL]

For what I know, The PCIe bus is a bottle neck. Using it to call textures from time to time may be possible, but using it to exchange geometry data to build bigger scenes would imho slow down the process. Each time you will have to move your scene in the 3d space, you will have to wait for the refresh of data. I think that we would lose the interactivity.

[acc24ex]

yep, it's a matter of programming choice - the amount of time the memory had to be swapped around would inhibit the render speed - but there was a mention of octane 2.0 utilizing RAM for textures ..
http://render.otoy.com/forum/viewtopic. ... 78#p178387

[riggles]

No, it's not mandatory for GPU renders. Redshift already has out-of-core capabilities for both geometry and textures.

[ROUBAL]

No, it's not mandatory for GPU renders. Redshift already has out-of-core capabilities for both geometry and textures.

But do they keep the same speed/interactivity as Octane ?

[riggles]

But do they keep the same speed/interactivity as Octane ?

That's hard to say, first because I don't use Redshift, and second because Octane's version of out-of-core won't be released for a while to compare. Redshift is faster than Octane in general, but it's a biased renderer, so that shouldn't be too surprising. I'm sure the performance is affected to at least some degree when you start accessing system RAM, but I don't know how much. The Redshift users I've chatted with, however, said it's out-of-core features was the selling point for them and they seem to love Redshift.