diff --git a/pytorch_lightning/models/trainer.py b/pytorch_lightning/models/trainer.py
index 4b6c151f5f..68ffe8c505 100644
--- a/pytorch_lightning/models/trainer.py
+++ b/pytorch_lightning/models/trainer.py
@@ -458,9 +458,10 @@ class Trainer(TrainerIO):
         # filter out the weights that were done on gpu so we can load on good old cpus
         self.optimizers = model.configure_optimizers()
 
+        model.cuda(self.data_parallel_device_ids[0])
+
         # run through amp wrapper
         if self.use_amp:
-            model.cuda(self.data_parallel_device_ids[0])
 
             # An example
             model, optimizers = amp.initialize(