diff --git a/dnn/src/cuda/handle.cpp b/dnn/src/cuda/handle.cpp
index 39bd56b3..f00006ea 100644
--- a/dnn/src/cuda/handle.cpp
+++ b/dnn/src/cuda/handle.cpp
@@ -52,7 +52,9 @@ HandleImpl::HandleImpl(megcoreComputingHandle_t comp_handle)
     // Get stream from MegCore computing handle.
     megdnn_assert(
             CUDNN_VERSION == cudnnGetVersion(),
-            "cudnn version mismatch: compiled with %d; detected %zu at runtime",
+            "cudnn version mismatch: compiled with %d; detected %zu at runtime, may "
+            "caused by customized environment, for example LD_LIBRARY_PATH on LINUX "
+            "and PATH on Windows!!",
             CUDNN_VERSION, cudnnGetVersion());
 #if CUDA_VERSION >= 10010
     megdnn_assert(
diff --git a/src/tensorrt/impl/tensorrt_runtime_opr.cpp b/src/tensorrt/impl/tensorrt_runtime_opr.cpp
index ef232d8c..f0849eb0 100644
--- a/src/tensorrt/impl/tensorrt_runtime_opr.cpp
+++ b/src/tensorrt/impl/tensorrt_runtime_opr.cpp
@@ -217,6 +217,12 @@ SymbolVarArray TensorRTRuntimeOpr::make(
         std::shared_ptr<nvinfer1::ICudaEngine> engine,
         std::shared_ptr<GpuAllocator> gpu_allocator, const SymbolVarArray& src,
         const OperatorNodeConfig& config) {
+    mgb_assert(
+            NV_TENSORRT_VERSION == getInferLibVersion(),
+            "TensorRT version mismatch: compiled with %d; detected %d at runtime , may "
+            "caused by customized environment, for example LD_LIBRARY_PATH on LINUX "
+            "and PATH on Windows!!",
+            NV_TENSORRT_VERSION, getInferLibVersion());
     VarNodeArray var_node_array = cg::to_var_node_array(src);
     auto tensor_rt_opr = std::make_unique<TensorRTRuntimeOpr>(
             std::move(engine), std::move(gpu_allocator), var_node_array, config);