diff --git a/cmake/external/mklml.cmake b/cmake/external/mklml.cmake
index 54826cedb871690a82b535ae3ed102600277c622..32a9368a9f639468a2144548118677aaddecbdf2 100644
--- a/cmake/external/mklml.cmake
+++ b/cmake/external/mklml.cmake
@@ -39,8 +39,10 @@ IF(WIN32)
     SET(MKLML_IOMP_LIB            ${MKLML_LIB_DIR}/libiomp5md.lib)
     SET(MKLML_SHARED_LIB          ${MKLML_LIB_DIR}/mklml.dll)
     SET(MKLML_SHARED_IOMP_LIB     ${MKLML_LIB_DIR}/libiomp5md.dll)
-ELSE()  
-    SET(MKLML_VER "mklml_lnx_${TIME_VERSION}" CACHE STRING "" FORCE)
+ELSE()
+    #TODO(intel-huying):
+    #  Now enable Erf function in mklml library temporarily, it will be updated as offical version later.
+    SET(MKLML_VER "VsErf_mklml_lnx_${TIME_VERSION}" CACHE STRING "" FORCE)
     SET(MKLML_URL "http://paddlepaddledeps.cdn.bcebos.com/${MKLML_VER}.tgz" CACHE STRING "" FORCE)
     SET(MKLML_LIB                 ${MKLML_LIB_DIR}/libmklml_intel.so)
     SET(MKLML_IOMP_LIB            ${MKLML_LIB_DIR}/libiomp5.so)
diff --git a/paddle/fluid/operators/activation_op.h b/paddle/fluid/operators/activation_op.h
index c7df3ea58a91579e35ff0d486516271a6daf054f..e8f5530b7887c19679dd1071805c5a66602493ba 100644
--- a/paddle/fluid/operators/activation_op.h
+++ b/paddle/fluid/operators/activation_op.h
@@ -11,6 +11,7 @@ limitations under the License. */
 
 #pragma once
 #include <glog/logging.h>
+#include <algorithm>
 #include <string>
 #include <unordered_set>
 #include <utility>
@@ -24,6 +25,7 @@ limitations under the License. */
 #include "paddle/fluid/framework/eigen.h"
 #include "paddle/fluid/framework/op_registry.h"
 #include "paddle/fluid/operators/detail/safe_ref.h"
+#include "paddle/fluid/operators/math/blas.h"
 #include "paddle/fluid/platform/float16.h"
 
 #ifdef PADDLE_WITH_MKLDNN
@@ -301,8 +303,28 @@ template <typename T>
 struct GeluFunctor : public BaseActivationFunctor<T> {
   template <typename Device, typename X, typename Out>
   void operator()(Device d, X x, Out out) const {
+// Because the execute or device context can not be deliver here, it keep the
+// marco for NVCC.
+#if defined(PADDLE_WITH_MKLML) && !defined(_WIN32) && !defined(__APPLE__) && \
+    !defined(__OSX__) && !defined(PADDLE_WITH_CUDA)
+    auto x_data = x.data();
+    auto out_data = out.data();
+    int n = std::min(x.size(), out.size());
+
+    std::memset(out_data, 0, n * sizeof(T));
+    math::CBlas<T>::AXPY(n, static_cast<T>(M_SQRT1_2), x_data, 1, out_data, 1);
+    math::CBlas<T>::VMERF(n, out_data, out_data, VML_LA);
+    for (int i = 0; i < n; i++) {
+      out_data[i] += static_cast<T>(1);
+    }
+    math::CBlas<T>::VMUL(n, x_data, out_data, out_data);
+    for (int i = 0; i < n; i++) {
+      out_data[i] *= static_cast<T>(0.5);
+    }
+#else
     auto temp = (x * static_cast<T>(M_SQRT1_2)).erf();
     out.device(d) = x * static_cast<T>(0.5) * (static_cast<T>(1) + temp);
+#endif
   }
 };
 
diff --git a/paddle/fluid/operators/math/blas.h b/paddle/fluid/operators/math/blas.h
index f67f57827bc03e134bf87edd5bf033adb5098916..ce8109f64d62b0d412419107881952f1b4ffc75e 100644
--- a/paddle/fluid/operators/math/blas.h
+++ b/paddle/fluid/operators/math/blas.h
@@ -184,6 +184,9 @@ class Blas {
   template <typename T>
   void VINV(int n, const T* a, T* y) const;
 
+  template <typename T>
+  void VMERF(int n, const T* a, T* y, int64_t mode) const;
+
  private:
   const DeviceContext& context_;
 };
@@ -290,6 +293,11 @@ class BlasT : private Blas<DeviceContext> {
     Base()->template VINV<T>(args...);
   }
 
+  template <typename... ARGS>
+  void VMERF(ARGS... args) const {
+    Base()->template VMERF<T>(args...);
+  }
+
  private:
   const Blas<DeviceContext>* Base() const {
     return static_cast<const Blas<DeviceContext>*>(this);
diff --git a/paddle/fluid/operators/math/blas_impl.h b/paddle/fluid/operators/math/blas_impl.h
index 972366bc093f4b7f0a090cf31213f75ccd89fd82..ba995dabecbfab8c4952bb7efeaa381f8078821a 100644
--- a/paddle/fluid/operators/math/blas_impl.h
+++ b/paddle/fluid/operators/math/blas_impl.h
@@ -123,6 +123,11 @@ struct CBlas<float> {
   static void VINV(ARGS... args) {
     platform::dynload::vsInv(args...);
   }
+
+  template <typename... ARGS>
+  static void VMERF(ARGS... args) {
+    platform::dynload::vmsErf(args...);
+  }
 };
 
 template <>
@@ -223,6 +228,11 @@ struct CBlas<double> {
   static void VINV(ARGS... args) {
     platform::dynload::vdInv(args...);
   }
+
+  template <typename... ARGS>
+  static void VMERF(ARGS... args) {
+    platform::dynload::vmdErf(args...);
+  }
 };
 
 #else
@@ -625,6 +635,19 @@ void Blas<DeviceContext>::VINV(int n, const T *a, T *y) const {
 #endif
 }
 
+template <>
+template <typename T>
+void Blas<platform::CPUDeviceContext>::VMERF(int n, const T *a, T *y,
+                                             int64_t mode) const {
+#ifdef PADDLE_WITH_MKLML
+  CBlas<T>::VMERF(n, a, y, mode);
+#else
+  for (int i = 0; i < n; ++i) {
+    y[i] = std::erf(a[i]);
+  }
+#endif
+}
+
 }  // namespace math
 }  // namespace operators
 }  // namespace paddle
diff --git a/paddle/fluid/platform/dynload/mklml.h b/paddle/fluid/platform/dynload/mklml.h
index a260cda49138580b209e647af459e9392d9f18f1..a5b846f500f3677188b170dda76c65047d628064 100644
--- a/paddle/fluid/platform/dynload/mklml.h
+++ b/paddle/fluid/platform/dynload/mklml.h
@@ -86,6 +86,8 @@ extern void* mklml_dso_handle;
   __macro(vdPowx);                  \
   __macro(vsInv);                   \
   __macro(vdInv);                   \
+  __macro(vmsErf);                  \
+  __macro(vmdErf);                  \
   __macro(MKL_Set_Num_Threads)
 
 MKLML_ROUTINE_EACH(DECLARE_DYNAMIC_LOAD_MKLML_WRAP);