I'm considering a GPU system for deep learning applications, mainly for training models with large datasets. So I'm not sure whether it makes sense to choose Nvidia NV-Link over more Tesla V100 graphics cards? Is NV-Link the bottleneck that a system with two Tesla V100 and NV-Link is "faster" compared to four Tesla V100 without NV-Link? As I have seen, the Nvidias GPU system DGX-1 and DGX-2 uses NVLink.