Nouveau Driver Picks Up SVM Support Via HMM

[coder]

I asked the Blender devs if they can start using OpenCL 2.x & shared virtual memory for Blender cycles, but I was told it's not worth the effort.

The natural follow-up is then to ask what benefit they'd expect and how much effort it would require. They might be right that if you look at the communication time, it's small enough relative to the total rendering time that it wouldn't yield a significant performance improvement.

The other argument for SVM might be to render from main memory, to enable rendering support for larger scenes on smaller GPUs. However, they might feel that would hurt performance too much, relative to staying entirely in GPU memory.

I don't know - I'm just speculating. So, go ahead and ask them what they mean and why they think so.