diff --git a/paddle/fluid/framework/details/fetch_op_handle.cc b/paddle/fluid/framework/details/fetch_op_handle.cc
index 6c8b8937ebe646042f71cb58cfbc2d32426a4e3c..fe14e3e91da34e5993a68d10a13b275bab576ce6 100644
--- a/paddle/fluid/framework/details/fetch_op_handle.cc
+++ b/paddle/fluid/framework/details/fetch_op_handle.cc
@@ -63,8 +63,7 @@ void FetchOpHandle::RunImpl() {
     auto &t = var->Get<framework::LoDTensor>();
     if (platform::is_gpu_place(t.place())) {
 #ifdef PADDLE_WITH_CUDA
-      TensorCopy(t, cpu, *dev_ctxes_.at(t.place()), &tensors_[i]);
-      dev_ctxes_.at(t.place())->Wait();
+      TensorCopy(t, cpu, &tensors_[i]);
 #endif
     } else {
       tensors_[i].ShareDataWith(t);