Re: Success - calling custom CUDA kernels from Nim

[Udiknedormin]

That's really good news. It means it should be 100% safe for programs not 
requiring more GPU memory than what your GPU can actually offer. For the ones 
that do, I would advise you to use regions or write RAII-like templates 
yourself as I don't think there is any way for Nim's GC to handle GPU memory 
(although maybe it's possible to write a separate GC implementation that does?).