GPU Out of Core memory possible or already planned?

[EDELVIZ_ARCH]

Hi Guys,

I am do huge landscape scenes with lots of displacement and tons of proxies. My scenes are usually using anywhere between 6 and 45 GB of ram. I currently only have very basic Graphics cards with 2GB of ram and am wondering if at some point I will be able to move to GPU. A collegue of mine is using redshift and said that with their auto proxy, auto tilable exr structure and out of core caching he can render whole landscapes in redshift on a two and even one gb card.
Since I am a complete GPU beginner, and currently dont have a powerful card to test V-Ray GPU I just wanted to ask in a noobish way if, a) V-Ray GPU has some out of core memory thing and if not, b) is it is planned? Is it realistic to expect to ever render a 45gb ram scene on a gpu?

Just to be clear, I am super happy with V-Ray, not complaining about it or comparing it to Redshift - just asking because that is my only informed point of reference.

Cheers,
Jonathan

[vlado]

If Redshift has to swap geometry out to disk, it tends to become quite slow to the point where advantages over CPU rendering might not be that apparent. Landscapes in particular do not require out-of-core rendering, as they are mostly instances.

To go back to your specific questions, we don't currently have out-of-core rendering for V-Ray GPU, but we do support instancing, and we have further plans to improve memory usage for V-Ray GPU, both when things fit in memory and when they don't. Also keep in mind that the GPU renderer generally uses less VRAM than system RAM, so even if your scenes take 45 GB for the CPU renderer, they will probably take up way less on the GPU.

Best regards,
Vlado

[GidPDX]

It's good to hear that more memory optimization is planned for RT. Vlado, do you know roughly what the ratio of needed system RAM vs. Vram is? I've heard many people say this, but I've never known what roughly the ratio is. I'm guessing it depends on what is being stored too? Images vs. mesh items etc.

Redshift DOES slow down when going out of core compared to when it can fit everything in Vram. It is not swapping to disc though, but RAM. That said, depending on the graphics card, the number of cards, the size of the scene and, very importantly, the CPU and MB you have, it is almost always significantly faster than CPU rendering for the same cost. This is something they've been working on since redshifts inception, so it's actually advanced along quite well.

Octane supports out of core textures. More out of core features are planned for octane as well, but otoy often says things are planned, but as to the actual date of being in a build...? They have a tendency to say things and not make good on them for a long time IMHO, so we shouldn't hold our breath.