pytorch failure Titan X (Maxwell)

elab@gpu5 ~/pytorch-examples/imagenet [master*]$ python3 main.py -a alexnet /media/SuperSSD/test-dataset-train-val/
=> creating model 'alexnet'
Traceback (most recent call last):
  File "main.py", line 286, in <module>
    main()
  File "main.py", line 129, in main
    train(train_loader, model, criterion, optimizer, epoch)
  File "main.py", line 165, in train
    output = model(input_var)
  File "/usr/local/lib/python3.4/dist-packages/torch/nn/modules/module.py", line 210, in __call__
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.4/dist-packages/torchvision/models/alexnet.py", line 42, in forward
    x = self.features(x)
  File "/usr/local/lib/python3.4/dist-packages/torch/nn/modules/module.py", line 210, in __call__
    result = self.forward(*input, **kwargs)
  File "/usr/local/lib/python3.4/dist-packages/torch/nn/parallel/data_parallel.py", line 41, in forward
    replicas = self.replicate(self.module, self.device_ids)
  File "/usr/local/lib/python3.4/dist-packages/torch/nn/parallel/data_parallel.py", line 48, in replicate
    return replicate(module, device_ids)
  File "/usr/local/lib/python3.4/dist-packages/torch/nn/parallel/replicate.py", line 33, in replicate
    param_copies = Broadcast(device_ids)(param)
  File "/usr/local/lib/python3.4/dist-packages/torch/nn/parallel/_functions.py", line 15, in forward
    return comm.broadcast(input, self.target_gpus)
  File "/usr/local/lib/python3.4/dist-packages/torch/cuda/comm.py", line 23, in broadcast
    nccl.broadcast(tensors)
  File "/usr/local/lib/python3.4/dist-packages/torch/cuda/nccl.py", line 182, in broadcast
    comm = communicator(inputs)
  File "/usr/local/lib/python3.4/dist-packages/torch/cuda/nccl.py", line 137, in communicator
    _communicators[key] = NcclCommList(devices)
  File "/usr/local/lib/python3.4/dist-packages/torch/cuda/nccl.py", line 110, in __init__
    check_error(lib.ncclCommInitAll(self, len(devices), int_array(devices)))
AttributeError: 'NoneType' object has no attribute 'ncclCommInitAll'
Exception ignored in: <bound method NcclCommList.__del__ of <torch.cuda.nccl.NcclCommList object at 0x7eff37a4f2e8>>
Traceback (most recent call last):
  File "/usr/local/lib/python3.4/dist-packages/torch/cuda/nccl.py", line 117, in __del__
    lib.ncclCommDestroy(self[i])
AttributeError: 'NoneType' object has no attribute 'ncclCommDestroy'