diff --git a/inferencer-fluid-gpu/include/fluid_gpu_engine.h b/inferencer-fluid-gpu/include/fluid_gpu_engine.h
index 314884b94deb0b27e2992484242d70c204585cfc..34c9a259f51f212733210a2bed7b0d79a80c8489 100644
--- a/inferencer-fluid-gpu/include/fluid_gpu_engine.h
+++ b/inferencer-fluid-gpu/include/fluid_gpu_engine.h
@@ -133,6 +133,7 @@ class FluidGpuAnalysisCore : public FluidFamilyCore {
     analysis_config.SetProgFile(data_path + "/__model__");
     analysis_config.EnableUseGpu(100, FLAGS_gpuid);
     analysis_config.SetCpuMathLibraryNumThreads(1);
+    analysis_config.EnableMemoryOptim(false, false);
     analysis_config.SwitchSpecifyInputNames(true);
     AutoLock lock(GlobalPaddleCreateMutex::instance());
     _core =
@@ -190,6 +191,7 @@ class FluidGpuAnalysisDirCore : public FluidFamilyCore {
     analysis_config.EnableUseGpu(100, FLAGS_gpuid);
     analysis_config.SwitchSpecifyInputNames(true);
     analysis_config.SetCpuMathLibraryNumThreads(1);
+    analysis_config.EnableMemoryOptim(false, false);
     AutoLock lock(GlobalPaddleCreateMutex::instance());
     _core =
         paddle::CreatePaddlePredictor<paddle::AnalysisConfig>(analysis_config);
@@ -497,6 +499,7 @@ class FluidGpuAnalysisDirWithSigmoidCore : public FluidGpuWithSigmoidCore {
     analysis_config.EnableUseGpu(100, FLAGS_gpuid);
     analysis_config.SwitchSpecifyInputNames(true);
     analysis_config.SetCpuMathLibraryNumThreads(1);
+    analysis_config.EnableMemoryOptim(false, false);
     AutoLock lock(GlobalPaddleCreateMutex::instance());
     _core->_fluid_core =
         paddle::CreatePaddlePredictor<paddle::AnalysisConfig>(analysis_config);