diff --git a/paddle/fluid/inference/tensorrt/engine.h b/paddle/fluid/inference/tensorrt/engine.h
index 421842cf563db0212f1bb0621fb7a819f9423d3f..aa5a7657e28cd5d52a957788eb6c0d9008711b5b 100644
--- a/paddle/fluid/inference/tensorrt/engine.h
+++ b/paddle/fluid/inference/tensorrt/engine.h
@@ -358,7 +358,9 @@ class TensorRTEngine {
   bool WithFp16() {
     bool enable_fp16 = (precision_ == AnalysisConfig::Precision::kHalf);
     bool support_fp16 = infer_builder_->platformHasFastFp16();
-    return enable_fp16 && support_fp16;
+    // below is consistent with setFlag in engine.cc
+    bool fall_back_fp16 = WithInt8() && !use_dla_;
+    return (enable_fp16 || fall_back_fp16) && support_fp16;
   }
 
   bool WithInt8() {