diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py
index 79aa1cb2..7b763136 100644
--- a/pytorch_lightning/models/trainer.py
+++ b/pytorch_lightning/models/trainer.py
@@ -96,7 +96,7 @@ class Trainer(TrainerIO):
             os.environ["CUDA_DEVICE_ORDER"] = "PCI_BUS_ID"
             os.environ["CUDA_VISIBLE_DEVICES"] = ','.join([str(x) for x in self.data_parallel_device_ids])
 
-        self.data_parallel = self.data_parallel_device_ids is not None and len(self.data_parallel_device_ids) > 0
+        self.data_parallel = self.data_parallel_device_ids is not None and len(self.data_parallel_device_ids) > 1
 
         # process info
         self.proc_rank = 0
@@ -295,7 +295,7 @@ class Trainer(TrainerIO):
 
         # when using gpus, first thing we do is spawn a new process between each worker
         # applies to single gpu, multi-gpu and multi-nodes
-        if self.on_gpu and len(self.data_parallel_device_ids) > 1:
+        if self.data_parallel:
             self.experiment = self.experiment.get_meta_copy()
             mp.spawn(self.dp_train, nprocs=len(self.data_parallel_device_ids), args=(model, ))
         else: