diff --git a/paddle/fluid/operators/reader/lod_tensor_blocking_queue.h b/paddle/fluid/operators/reader/lod_tensor_blocking_queue.h
index 17778148f82dd70ec0ed6d7b1dccfc43e1550972..19f564946f903440af4513ea3b0db462455e5669 100644
--- a/paddle/fluid/operators/reader/lod_tensor_blocking_queue.h
+++ b/paddle/fluid/operators/reader/lod_tensor_blocking_queue.h
@@ -74,7 +74,9 @@ class OrderedMultiDeviceLoDTensorBlockingQueue {
  public:
   OrderedMultiDeviceLoDTensorBlockingQueue(size_t capacity,
                                            bool speed_test_mode = false)
-      : capacity_(capacity), speed_test_mode_(speed_test_mode) {}
+      : capacity_(capacity), speed_test_mode_(speed_test_mode) {
+    LOG(INFO) << "Use keep_order=True DataLoader";
+  }
 
   ~OrderedMultiDeviceLoDTensorBlockingQueue() {
     VLOG(10) << "Destruct OrderedMultiDeviceLoDTensorBlockingQueue";
diff --git a/python/paddle/fluid/reader.py b/python/paddle/fluid/reader.py
index 67e6259e8124f8898ef6c53a880ec3ea2505be31..ed6ac17e1b116b22048cb87167385e3d86b97d95 100644
--- a/python/paddle/fluid/reader.py
+++ b/python/paddle/fluid/reader.py
@@ -89,7 +89,7 @@ class DataLoader(object):
                        return_list=False,
                        use_multiprocess=False,
                        drop_last=True,
-                       keep_order=True):
+                       keep_order=False):
         """
         Create a DataLoader object for loading data from Python generator. 
         Data would be prefetched using Python thread and be pushed
@@ -633,7 +633,7 @@ class GeneratorLoader(DataLoaderBase):
                  iterable=True,
                  return_list=False,
                  drop_last=True,
-                 keep_order=True):
+                 keep_order=False):
         self._tensor_reader = None
         self._places = None
         self._thread = None