Is it possible to build acceleration structures asynchronously so when a bottom level structure is marked dirty it does not introduce delay to the render time ? Would it be possible to build the structure in CPU and upload the results to GPU when they are done ?
No, the OptiX API is not thread safe.
See caveats in the OptiX Programming Guide: [url]http://raytracing-docs.nvidia.com/optix/guide/index.html#caveats#caveats[/url]