diff --git a/paddle/fluid/framework/details/fetch_op_handle.cc b/paddle/fluid/framework/details/fetch_op_handle.cc
index 946ee91a667496e2427304df4228334bb1061890..423449abff97dbf70d81314f852d9135e25f243f 100644
--- a/paddle/fluid/framework/details/fetch_op_handle.cc
+++ b/paddle/fluid/framework/details/fetch_op_handle.cc
@@ -66,7 +66,7 @@ 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_[t.place()], &tensors_[i]);
+      TensorCopy(t, cpu, *dev_ctxes_[t.place()], &tensors_[i], true);
       dev_ctxes_.at(t.place())->Wait();
 #endif
     } else {