The system requirements for Parallel Computing Toolbox are describedhere.
You need an NVIDIA GPU with compute capability at least 1.3 (1.3 is required to give double-precision support, and also numeric consistency with MATLAB). NVIDIA have a listing of their GPUs with CUDA supporthere.
You can't build CUDA kernels usingmex without Parallel Computing Toolbox. This requires the NVIDIA compiler nvcc and the infrastructure to call that frommex is only provided with PCT.
You need to build your device code into a separate library independently of MATLAB and built usingnvcc, and then write mex functions that link to it using the-l option.
Best Answer