Nvidia NVlink Resource Pooling?

I’ve been researching for building a personal render farm/new primary work station, and I’m wondering how extensive Blender’s support is for NVIDIA NVlink, specifically for rendering in Cycles.

Some of the Quadro series cards support memory pooling through NVlink, with some rumors pointing towards 2080ti cards supporting it. It does however depend on the software support for NVlink.

My main question, is if Blender currently supports (2.81) or if it will support resource pooling for cycles?
This way you could use dual Quadro RTX 6000 cards with NVlink to support memory pooling, creating a 48GB memory buffer for Cycles.

This combined with the latest support for Optix, and speed improvements could mean much higher render speeds, combined with higher VRAM for a fair price.

So in summary, does Blender and Cycles support Resource and VRAM pooling with NVlink?
Doe it exist? Is it Possible? If it’s not possible, I’d love to learn more information on the technical side of the support for such a feature, and why it might not be possible to implement for Ray Traced render engines.

Thanks for any input and thoughts.

~ Cheers

I am not aware of something like that nor plans for. Currently it works, that a copy of the scene is loaded to each card and the only thing in terms of coordination is that the cards render different tiles of the frame.

Not Cycles, but it is supported in Octane. It only kicks in if you run out of vram on a single card as NVlink does slow the rendering down a bit. It’s a lot like using out-of-core system memory.

I’ll leave this here, informatively.
https://www.techpowerup.com/261357/nvidia-develops-tile-based-multi-gpu-rendering-technique-called-cfr