Hi @ptrblck ! I’m very new to Nvlink, do we need to write any Initial code to enable the usage of nvlink while running our model and could you please give me any reference example code of a sample model which is running on nvlink for data parallelism using nvlink, if you have any.
Thank you
NVLink should be used automatically for all device2device communications.
You can check the status via nvidia-smi nvlink -s and see the connectivity matrix via nvidia-smi topo -m.
While the first command should give you the current status and bandwidth, the latter one will show you shich GPUs are connected to each other with which technology.