Hi there,
I’m working on scenes with lots of 4k materials, and I’m struggling with optimization for the whole thing to fit in the memory of my 8 GB RTX 2080. I know we can now use the system memory (OOC) with the GPU but I’m still limited to 32GB, which is not so much when the scene is loaded in 3dsmax.
I see that EXR or TX tiled textures are very efficient with the CPU (I have plenty of 4k materials in the background), but unfortunately it doesn’t seem to work with the GPU engine 
These days, GPUs are so expensive, it would be awesome if we could have the possibility to optimize the VRAM costs with this elegant solution.
I’ve seen on a Blender forum that the feature could be tough to implement for GPU engines, is it the reason why it doesn’t work with Vray GPU for now? Could the out-of-core textures option help in this case (I mean, help the GPU to use the tiled textures located in the system RAM)?
Thanks for reading 
Try the suggestions from this article.
Thank you! Actually I already tried all of these tips (the recap is really useful!), I’m currently using the resize feature to clamp my texture size to 2k, with GI set to full brute force. It works well for distant materials, but I lose details on near objects and large surfaces, obviously.
Having the possibility to use tiled or mipmapped textures would allow to use full texture size, while maintaining very low render times.
The thing is, we used to have a “mipmapping” texture mode for the GPU, but it didn’t work well, and OOC does the same job, but better, hence why we removed it.
̶O̶n̶e̶ ̶o̶t̶h̶e̶r̶ ̶t̶h̶i̶n̶g̶ ̶t̶h̶a̶t̶ ̶c̶o̶u̶l̶d̶ ̶b̶e̶ ̶v̶e̶r̶y̶ ̶h̶e̶l̶p̶f̶u̶l̶ ̶(̶I̶ ̶s̶h̶o̶u̶l̶d̶ ̶a̶d̶d̶ ̶i̶t̶ ̶t̶o̶ ̶t̶h̶e̶ ̶a̶r̶t̶i̶c̶l̶e̶)̶ ̶i̶s̶ ̶t̶o̶ ̶u̶s̶e̶ ̶t̶h̶e̶ ̶N̶V̶i̶d̶i̶a̶ ̶A̶I̶ ̶U̶p̶s̶c̶a̶l̶e̶r̶ ̶(̶l̶o̶c̶a̶t̶e̶d̶ ̶i̶n̶ ̶t̶h̶e̶ ̶V̶R̶a̶y̶D̶e̶n̶o̶i̶s̶e̶r̶ ̶o̶p̶t̶i̶o̶n̶s̶)̶.̶ ̶I̶t̶ ̶w̶o̶u̶l̶d̶ ̶c̶e̶r̶t̶a̶i̶n̶l̶y̶ ̶h̶e̶l̶p̶ ̶o̶u̶t̶,̶ ̶s̶i̶n̶c̶e̶ ̶i̶n̶t̶e̶r̶n̶a̶l̶l̶y̶ ̶t̶h̶e̶ ̶i̶m̶a̶g̶e̶ ̶w̶i̶l̶l̶ ̶b̶e̶ ̶r̶e̶n̶d̶e̶r̶e̶d̶ ̶a̶t̶ ̶a̶ ̶l̶o̶w̶e̶r̶ ̶r̶e̶s̶o̶l̶u̶t̶i̶o̶n̶,̶ ̶t̶h̶u̶s̶ ̶s̶a̶v̶i̶n̶g̶ ̶m̶e̶m̶o̶r̶y̶.̶.
EDIT: The devs told me that this won’t help with memory optimization, only speed. GPU doesn’t buffer the whole image, and CPU always buffers the original resolution.
Ok, I’ll stick to the OOC option for the moment, even if render times are higher than without using it, it’s still very fast compared to the CPU only render
And I’m going to check out the upscaler… Thank you!