diff --git a/pytorch_lightning/plugins/training_type/ddp.py b/pytorch_lightning/plugins/training_type/ddp.py
index 220a18c4c9b295..a13428c51f09b5 100644
--- a/pytorch_lightning/plugins/training_type/ddp.py
+++ b/pytorch_lightning/plugins/training_type/ddp.py
@@ -326,7 +326,7 @@ def post_dispatch(self) -> None:
     def barrier(self, *args, **kwargs) -> None:
         if not torch_distrib.is_initialized():
             return
-        if _TORCH_GREATER_EQUAL_1_8 and self.torch_distributed_backend == "nccl":
+        if _TORCH_GREATER_EQUAL_1_8 and torch_distrib.dist_backend == "nccl":
             torch_distrib.barrier(device_ids=self.determine_ddp_device_ids())
         else:
             torch_distrib.barrier()
diff --git a/pytorch_lightning/plugins/training_type/ddp_spawn.py b/pytorch_lightning/plugins/training_type/ddp_spawn.py
index cd0ecb5f4039e4..323fa19dab82f9 100644
--- a/pytorch_lightning/plugins/training_type/ddp_spawn.py
+++ b/pytorch_lightning/plugins/training_type/ddp_spawn.py
@@ -312,8 +312,7 @@ def __recover_child_process_weights(self, best_path, last_path):
     def barrier(self, *args, **kwargs) -> None:
         if not torch_distrib.is_initialized():
             return
-        if _TORCH_GREATER_EQUAL_1_8 and self.torch_distributed_backend == "nccl":
-            assert torch_distrib.dist_backend == "nccl"
+        if _TORCH_GREATER_EQUAL_1_8 and torch_distrib.dist_backend == "nccl":
             torch_distrib.barrier(device_ids=self.determine_ddp_device_ids())
         else:
             torch_distrib.barrier()