Closed Qubitium closed 1 year ago
CMAKE_CUDA_ARCHITECTURES
is a CMake option documented here: https://cmake.org/cmake/help/latest/variable/CMAKE_CUDA_ARCHITECTURES.html
It is used as the default value for CUDA_ARCHITECTURES
on a CUDA target.
This option is akin to the architecture specification options in nvcc: https://docs.nvidia.com/cuda/cuda-compiler-driver-nvcc/#base-notation
You can look up the compute capability of a GPU here: https://developer.nvidia.com/cuda-gpus
So for your example of a 3090, the compute capability is 8.6. So to build both PTX and SASS for that GPU you would specify -DCMAKE_CUDA_ARCHITECTURES=86
.
A nice feature starting with CMake 3.24 is the ability to say -DCMAKE_CUDA_ARCHITECTURES=native
which will automatically detect the GPU on your system and build for that GPU.
What are the values for
CMAKE_CUDA_ARCHITECTURES
? For example, for Ampere 3090 series.Cannot find this value anywhere. Thanks.