CUDA version

I am using Optix 3.0.1 and Optix.1.dll from the toolkit depends on cudart32_42_9_.dll.
The release notes for this version of Optix say: “OptiX 3.0.1 has been built with CUDA 4.2 and 5.0, but any specified toolkit should work when compiling PTX for OptiX …etc” regarding PTX compilation. So, despite the PTX files are compiled with CUTA 5.0 compiler (for example), CUDA 4.2 is actually used.
I have 2 questions.

  1. Is it possible to use another CUDA version not only for PTX compilation, but in runtime also?
  2. (the problen whch I am trying to solve) The application is using Optix with CUDA 4.2 and another Dll with CUDA 5.0. The functions that call GPU work sequentually (CUDA 4.2 → CUDA 5.0 → end). Running both functions, the program does not fail, it counts the result, but never finishes. Separately, they are working fine. Can it be caused by using 2 different versions of CUDA?