diff --git a/paddle/fluid/operators/math/concat_and_split.cu b/paddle/fluid/operators/math/concat_and_split.cu
index 760a065c1081d1e55901774b258ba524471b856b..930d851696efdf8dc92f98638f2d09cac7eb304e 100644
--- a/paddle/fluid/operators/math/concat_and_split.cu
+++ b/paddle/fluid/operators/math/concat_and_split.cu
@@ -180,7 +180,7 @@ class ConcatFunctor<platform::CUDADeviceContext, T> {
     }
     // Wait() must be called because `inputs_data` may be destructed before
     // kernel ends
-    context.Wait();
+    /* context.Wait(); */
   }
 };
 
@@ -258,7 +258,7 @@ class SplitFunctor<platform::CUDADeviceContext, T> {
     }
     // Wait() must be called because `outputs_data` may be destructed before
     // kernel ends
-    context.Wait();
+    /* context.Wait(); */
   }
 };