Card support and different optix / cuda versions

I have a question about which cards are supported in which versions.
The following two points have been very clearly communicated throughout:

  • Optix 3.8 / CUDA 7.0 - Fermi, Kepler & Maxwell.
  • Optix 4.0.2 / CUDA 7.5 - Kepler, Maxwell & Pascal.

My question is about the two in-between versions, 3.9 & 3.9.1


  • 3.9 Release Notes: "CUDA capable devices of Compute Capability 2.0 (“Fermi”) or higher are supported on GeForce, Quadro, or Tesla class NVIDIA products."
  • 4.0.0 Release Notes: "Support for Fermi class GPUs has been removed."
  • 4.0.2 Release Notes: "All NVIDIA GPUs of Compute Capability 3.0 (Kepler) or higher are supported. Note that this is a change from OptiX 3.9, as Fermi class GPUs are no longer supported."


So here comes my actual question: Am I understanding correctly, that version 3.9.1 supports both Pascal and Fermi architectures? If yes, does it matter if I use CUDA sdk 7.0 or 7.5 to compile my code (in and for Windows only)?


OptiX 3.9.1 used with CUDA skd 7.5 supports both Pascal and Fermi architectures. I’m not sure about 7.0.

The CUDA requirement applies only to the driver on the system on which the optix-based application is run. The CUDA toolkit version used to compile input PTX for the optix API is basically unrestricted. After all we want old optix-based applications (with pre-compiled PTX) to be able to drop in a newer optix lib and still run.