New laptop with GTX 1060. Just installed Ubuntu 16.04 and Cuda Toolkit 8.0. I start by creating a new project: Cuda C / C ++ Project-> CUDA Runtime Projekt -> specify a name, etc.
Then we move on to setting up the Generate PTX / GPU code. If I select the standard 3.5 and create a project and run the generated example, everything works as expected.
But if I choose only 6.0 to generate the PTX and GPU code, the sample project will still run, but I will not have any cuda syntax or autocompletion in the file. Adding certain header files, such as cuda_runtime, etc., does not fix it.
I thought this parameter was somewhat reminiscent of "Set Computing Ability XX" to get all the new cuda features with a higher version (like shfl, dynamic parallelism, etc.). That's why I thought it should work, like everything else, if I choose 6.0 instead of 3.5.
Am I missing something or am I just mistaken in the expected behavior?
source
share