Hope this is the right forum…
I’ve installed the CUDA_8.0.61 kit for Windows 7.
Compiling and debugging with Visual Studio 2015.
When I call cudaMallocManaged() it returns “cudaErrorNotSupported”.
My NVidia adapter is a Quadro 2000M.
I am compiling for x64.
What could the problem be?
OK - it appears my device does not support Unified memory.
So there must be an alternative?
Quadro 2000M is a Fermi-family device, with compute capability 2.1. It should be supported by CUDA 8. I am reasonably sure that cudaMallocManaged requires compute capability >= 3.0.
OK. I see from examples, that for this level of compute, I need to do a cudaMalloc and a cudaMemcpy to get data back and forth from Host to Device.