Any pointers on memory management for large scenes?
I am doing a lot...
All textures are tiled 8bit .tx files or tiled .exrs when float is required. All VRayHDRI nodes, not bitmaps.
High poly models are vRay proxies whenever possible.
I am using embree for motion blur (does that ignore the BSP settings completely?) I tried traditional motion blur but got higher memory usage (possibly was too aggressive with the face level coefficient). I have enabled conserve memory for embree.
My objects tend to animate and then go to visibility 0. The shot has lots of pieces transforming into other things. Flipping around and turning on an off.
Is animating the visibility (which typically has a stepped curve in my case) the most efficient method for this? Will geometry that is visibility 0 still use memory and slow things down? (Seem like it may be). Is there a better method? Delete mesh modifier? If a vRay proxy has a visibilty of 0 is it ever loaded? Hopefully not... .?
About 8 to 10 of our render nodes still have only 12GB of ram, and they are thrashing on this. Trying to get their speed back be reducing the memory footprint.
My understanding is that if I have filter maps for GI turned on then the GI will only use the lower res MIP maps as needed from the tiled images... Right? I have a custom GI filtering strength of 2.2 as it seems to reduce noise in the interior. Does this custom strength mean the maps get loaded twice?
The light cache eats up a lot of memory. Any way to help with that other than fewer samples? I have to use camera path or I get flicker.
Any thoughts, tips?
Thanks.
I am doing a lot...
All textures are tiled 8bit .tx files or tiled .exrs when float is required. All VRayHDRI nodes, not bitmaps.
High poly models are vRay proxies whenever possible.
I am using embree for motion blur (does that ignore the BSP settings completely?) I tried traditional motion blur but got higher memory usage (possibly was too aggressive with the face level coefficient). I have enabled conserve memory for embree.
My objects tend to animate and then go to visibility 0. The shot has lots of pieces transforming into other things. Flipping around and turning on an off.
Is animating the visibility (which typically has a stepped curve in my case) the most efficient method for this? Will geometry that is visibility 0 still use memory and slow things down? (Seem like it may be). Is there a better method? Delete mesh modifier? If a vRay proxy has a visibilty of 0 is it ever loaded? Hopefully not... .?
About 8 to 10 of our render nodes still have only 12GB of ram, and they are thrashing on this. Trying to get their speed back be reducing the memory footprint.
My understanding is that if I have filter maps for GI turned on then the GI will only use the lower res MIP maps as needed from the tiled images... Right? I have a custom GI filtering strength of 2.2 as it seems to reduce noise in the interior. Does this custom strength mean the maps get loaded twice?
The light cache eats up a lot of memory. Any way to help with that other than fewer samples? I have to use camera path or I get flicker.
Any thoughts, tips?
Thanks.
Comment