diff --git a/invokeai/backend/model_management/model_cache.py b/invokeai/backend/model_management/model_cache.py
index b4db4cf5ea..12d718ff86 100644
--- a/invokeai/backend/model_management/model_cache.py
+++ b/invokeai/backend/model_management/model_cache.py
@@ -265,11 +265,13 @@ class ModelCache(object):
 
         return self.ModelLocker(self, key, cache_entry.model, gpu_load, cache_entry.size)
 
-    def _move_model_to_device(self, key, target_device):
+    def _move_model_to_device(self, key: str, target_device: torch.device):
         cache_entry = self._cached_models[key]
 
         source_device = cache_entry.model.device
-        if source_device == target_device:
+        # Note: We compare device types only so that 'cuda' == 'cuda:0'. This would need to be revised to support
+        # multi-GPU.
+        if source_device.type == target_device.type:
             return
 
         start_model_to_time = time.time()