diff --git a/paddle/fluid/inference/api/analysis_config.cc b/paddle/fluid/inference/api/analysis_config.cc
index 2189b873818296d7668dc73549dea8251e4c5755..a9e477f883efc60793fc90e826da7d0e1cada13d 100644
--- a/paddle/fluid/inference/api/analysis_config.cc
+++ b/paddle/fluid/inference/api/analysis_config.cc
@@ -205,8 +205,6 @@ void AnalysisConfig::Update() {
       // Append after the Affine_channel_conv_fuse pass.
       pass_builder()->InsertPass(3, "tensorrt_subgraph_pass");
     }
-    // runtime_context_cache isn't fit for tensorrt.
-    enable_runtime_context_cache_ = false;
   }
 
   if (use_mkldnn_) {
diff --git a/paddle/fluid/inference/api/paddle_analysis_config.h b/paddle/fluid/inference/api/paddle_analysis_config.h
index 85639eebe4dd2eb713e1637b584e139bd0e22eef..3b7faa54000a26310d0117fe6f1e68cc404c461a 100644
--- a/paddle/fluid/inference/api/paddle_analysis_config.h
+++ b/paddle/fluid/inference/api/paddle_analysis_config.h
@@ -278,7 +278,7 @@ struct AnalysisConfig {
   // since the input/output names of this Op do not change in the execution,
   // RuntimeContext could be created only at the first iteration of this Op's
   // execution to save the elapsed time.
-  bool enable_runtime_context_cache_{true};
+  bool enable_runtime_context_cache_{false};
 
   // A runtime cache, shouldn't be transferred to others.
   std::string serialized_info_cache_;
diff --git a/paddle/fluid/inference/tests/api/analyzer_pyramid_dnn_tester.cc b/paddle/fluid/inference/tests/api/analyzer_pyramid_dnn_tester.cc
index 5157bd280d0f3ee327d5cee7799477b5e6fd3f71..e1787a71775207d4d94f4005cffb82c2b24274e6 100644
--- a/paddle/fluid/inference/tests/api/analyzer_pyramid_dnn_tester.cc
+++ b/paddle/fluid/inference/tests/api/analyzer_pyramid_dnn_tester.cc
@@ -107,6 +107,7 @@ void SetConfig(AnalysisConfig *cfg) {
   cfg->DisableGpu();
   cfg->SwitchSpecifyInputNames();
   cfg->SwitchIrOptim();
+  cfg->SwitchRuntimeContextCache();
   if (FLAGS_zero_copy) {
     cfg->SwitchUseFeedFetchOps(false);
   }