diff --git a/dnn/include/megdnn/common.h b/dnn/include/megdnn/common.h
new file mode 100644
index 0000000000000000000000000000000000000000..93fa158dc6c1d36c1054ced407f4ecf968cf87e3
--- /dev/null
+++ b/dnn/include/megdnn/common.h
@@ -0,0 +1,26 @@
+/**
+ * \file dnn/include/megdnn/common.h
+ * MegEngine is Licensed under the Apache License, Version 2.0 (the "License")
+ *
+ * Copyright (c) 2014-2021 Megvii Inc. All rights reserved.
+ *
+ * Unless required by applicable law or agreed to in writing,
+ * software distributed under the License is distributed on an
+ * "AS IS" BASIS, WITHOUT ARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ */
+
+#pragma once
+
+#if MGB_ENABLE_GETENV
+#define MGB_GETENV  ::std::getenv
+#else
+#define MGB_GETENV(_name)  static_cast<char*>(nullptr)
+#endif
+
+#ifdef WIN32
+#define unsetenv(_name) _putenv_s(_name, "");
+#define setenv(name,value,overwrite) _putenv_s(name,value)
+#endif
+
+
+// vim: syntax=cpp.doxygen foldmethod=marker foldmarker=f{{{,f}}}
diff --git a/dnn/src/cuda/handle.cpp b/dnn/src/cuda/handle.cpp
index 2c6eb987d2aa56afeb33e1cddce751787b9c8919..3ab550941cc67ebad6ccc89f72a79b18b3f99964 100644
--- a/dnn/src/cuda/handle.cpp
+++ b/dnn/src/cuda/handle.cpp
@@ -16,6 +16,7 @@
 #include "src/cuda/handle.h"
 #include "src/cuda/utils.h"
 #include "src/cuda/api_cache.h"
+#include "megdnn/common.h"
 
 #include <cuda.h>
 #include <cstring>
@@ -59,10 +60,12 @@ HandleImpl::HandleImpl(megcoreComputingHandle_t comp_handle):
         "cuda library version is too low to run cublasLt");
 #endif
 #if CUDNN_VERSION >= 8000
-    megdnn_log_warn(R"(
-        Cudnn8 will jit ptx code with cache. You can set 
-        CUDA_CACHE_MAXSIZE and CUDA_CACHE_PATH environment var to avoid repeat jit(very slow).
-        For example `export CUDA_CACHE_MAXSIZE=2147483647` and `export CUDA_CACHE_PATH=/data/.cuda_cache`)");
+    if (!MGB_GETENV("CUDA_CACHE_PATH")) {
+        megdnn_log_warn(R"(
+            Cudnn8 will jit ptx code with cache. You can set 
+            CUDA_CACHE_MAXSIZE and CUDA_CACHE_PATH environment var to avoid repeat jit(very slow).
+            For example `export CUDA_CACHE_MAXSIZE=2147483647` and `export CUDA_CACHE_PATH=/data/.cuda_cache`)");
+    }
 #endif
     cudnn_check(cudnnCreate(&m_cudnn_handle));
     cublas_check(cublasCreate(&m_cublas_handle));
diff --git a/src/core/include/megbrain/common.h b/src/core/include/megbrain/common.h
index e999515e189be124e0acdb97dbe701ea465fd8ac..5283e1b8188f05dd3b535dc245535b35a23d035e 100644
--- a/src/core/include/megbrain/common.h
+++ b/src/core/include/megbrain/common.h
@@ -13,6 +13,7 @@
 
 #include "megbrain_build_config.h"
 #include "megdnn/basic_types.h"
+#include "megdnn/common.h"
 
 #include <memory>
 #include <string>
@@ -199,17 +200,6 @@ void __log__(LogLevel level, const char *file, const char *func, int line,
 
 /* ================ misc ================  */
 
-#if MGB_ENABLE_GETENV
-#define MGB_GETENV  ::std::getenv
-#else
-#define MGB_GETENV(_name)  static_cast<char*>(nullptr)
-#endif
-
-#ifdef WIN32
-#define unsetenv(_name) _putenv_s(_name, "");
-#define setenv(name,value,overwrite) _putenv_s(name,value)
-#endif
-
 // use some macro tricks to get lock guard with unique variable name
 #define MGB_TOKENPASTE(x, y) x ## y
 #define MGB_TOKENPASTE2(x, y) MGB_TOKENPASTE(x, y)