Picking the right version of the NVIDIA CUDA driver can be the most difficult part of the CUDA install. If you choose the wrong version, the resulting error messages may not always be clear. This guide will help you to choose the right CUDA version for your system. It may also be necessary to change your OptiX version if using an old GPU along with the radiation model. Fortunately, this is easy and will also be described below.
There are three main considerations when choosing your CUDA version:
The system requirements for any CUDA version can be found in the release notes, which is both packaged with the CUDA toolkit and can be found on the page where you download CUDA.
The table below gives recommendations for Windows and Linux Systems based on your compute capability.
| Operating System | Compute capability | Recomm. Compiler | Recomm. CUDA | Use legacy OptiX? |
|---|---|---|---|---|
| Windows | 3.5 | Visual Studio 2019 | 10.2 | Yes |
| 5.0+ | Visual Studio 2022 | Latest | No | |
| Linux | 3.5 | gcc/g++ 6.0-8.0 | 10.2 | Yes |
| 5.0+ | gcc/g++ 7.0-11.0 | Latest | No |
The NVIDIA OptiX library is used for ray-tracing calculations in the radiation model. Helios bundles three versions of OptiX and selects among them automatically based on your NVIDIA driver version:
| Driver version | OptiX version used | CUDA requirement |
|---|---|---|
| ≥ 560 | OptiX 8.1 (default, recommended) | CUDA 12.0+ |
| < 560 | OptiX 6.5 (legacy) | CUDA 9.0+ |
| Any (or no NVIDIA GPU) | Vulkan compute backend | None |
Legacy GPUs (compute capability 3.5): If you are using a GPU with compute capability < 5.0, you can force the OptiX 6.5 legacy backend by setting -DOPTIX_VERSION_LEGACY=ON. There are two ways to do this:
set( OPTIX_VERSION_LEGACY ON ) to your project CMakeLists.txt file.