Comment by ZeroCool2u

Seems almost too good to be true, but I really hope it's not. How does it handle things like CUDA dependencies? Can it somehow make those portable too? Or is GPU acceleration not quite there yet?