Disclaimer: there is a lot of gap in my knowledge about this question.
Context: In the past I was using Optix for nearest neighbor search (I promise that in my context using raycasting for approximate nearest neighbor was not as dumb as it sounds), then migrated to cubql for various reasons. This is a shared GPU on a robot, so there are mild real time constraints and GPU sharing issues.
My understanding is that Optix makes use of RTX, and that Optix under the hood builds a BVH just like cuBQL. So I was wondering whether cuBQL used RTX too.
When I used Optix instead of cuBQL, one benefit was that on our shared GPU my process was the only process using RTX, and therefore it was not conflicting with anything else. After switching to cuBQL, it seems that we have a bit more conflicts.
Does that make sense? Also is RTX used for the BVH building step or the query step or both?
Disclaimer: there is a lot of gap in my knowledge about this question.
Context: In the past I was using Optix for nearest neighbor search (I promise that in my context using raycasting for approximate nearest neighbor was not as dumb as it sounds), then migrated to cubql for various reasons. This is a shared GPU on a robot, so there are mild real time constraints and GPU sharing issues.
My understanding is that Optix makes use of RTX, and that Optix under the hood builds a BVH just like cuBQL. So I was wondering whether cuBQL used RTX too.
When I used Optix instead of cuBQL, one benefit was that on our shared GPU my process was the only process using RTX, and therefore it was not conflicting with anything else. After switching to cuBQL, it seems that we have a bit more conflicts.
Does that make sense? Also is RTX used for the BVH building step or the query step or both?