diff --git a/ptypy/accelerate/cuda_cupy/multi_gpu.py b/ptypy/accelerate/cuda_cupy/multi_gpu.py
index 79f51142..6ccbce9a 100644
--- a/ptypy/accelerate/cuda_cupy/multi_gpu.py
+++ b/ptypy/accelerate/cuda_cupy/multi_gpu.py
@@ -98,7 +98,10 @@ def __init__(self):
         # get a unique identifier for the NCCL communicator and
         # broadcast it to all MPI processes (assuming one device per process)
         if self.rank == 0:
-            self.id = nccl.get_unique_id()
+            try:
+                self.id = nccl.get_unique_id()
+            except:
+                self.id = None
         else:
             self.id = None