diff --git a/paddle/fluid/inference/api/analysis_predictor.cc b/paddle/fluid/inference/api/analysis_predictor.cc
index 6388cfc4b2dae..97fde22308003 100644
--- a/paddle/fluid/inference/api/analysis_predictor.cc
+++ b/paddle/fluid/inference/api/analysis_predictor.cc
@@ -1054,7 +1054,6 @@ std::unique_ptr<PaddlePredictor> CreatePaddlePredictor<
                            std::to_string(fraction_of_gpu_memory);
         VLOG(3) << "set flag: " << flag;
         gflags.push_back(flag);
-        gflags.push_back("--cudnn_deterministic=True");
       }
 
 // TODO(wilber): jetson tx2 may fail to run the model due to insufficient memory