diff --git a/paddle/fluid/inference/api/analysis_predictor.cc b/paddle/fluid/inference/api/analysis_predictor.cc
index 8f51613d7f4fc5aefe1ddf7a8ad2c6863268946b..81c68a65576ca4cbfd58e915cae58f796115d1fe 100644
--- a/paddle/fluid/inference/api/analysis_predictor.cc
+++ b/paddle/fluid/inference/api/analysis_predictor.cc
@@ -653,6 +653,13 @@ std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<
         process_level_allocator_enabled = true;
       }
 
+// TODO(wilber): jetson tx2 may fail to run the model due to insufficient memory
+// under the native_best_fit strategy. Modify the default allocation strategy to
+// auto_growth. todo, find a more appropriate way to solve the problem.
+#ifdef WITH_NV_JETSON
+      gflags.push_back("--allocator_strategy=auto_growth");
+#endif
+
       if (framework::InitGflags(gflags)) {
         VLOG(3) << "The following gpu analysis configurations only take effect "
                    "for the first predictor: ";