diff --git a/src/operator/rnn-inl.h b/src/operator/rnn-inl.h
index 74a1aff5093f..60cc16deebd4 100644
--- a/src/operator/rnn-inl.h
+++ b/src/operator/rnn-inl.h
@@ -1928,6 +1928,60 @@ static void RNNStatefulComputeCPU(const OpStatePtr& state_ptr,
     op.Forward(ctx, in_blobs, req, out_blobs);
   });
 }
+
+static void RNNStatefulGradComputeCPU(const OpStatePtr& state_ptr,
+                                      const OpContext& ctx,
+                                      const std::vector<NDArray>& inputs,
+                                      const std::vector<OpReqType>& req,
+                                      const std::vector<NDArray>& outputs) {
+  std::vector<TBlob> in_blobs;
+  std::vector<TBlob> out_blobs;
+  std::vector<NDArray> temp_ndarrays_i;
+  std::vector<NDArray> temp_ndarrays_o;
+  for (const NDArray& in : inputs) {
+    if (in.storage_type() == kDefaultStorage) {
+      temp_ndarrays_i.push_back(in.Reorder2Default());
+      in_blobs.emplace_back(temp_ndarrays_i.back().data());
+    } else {
+      in_blobs.emplace_back(in.data());
+    }
+  }
+  for (const NDArray& out : outputs) {
+    if (out.storage_type() == kDefaultStorage) {
+      temp_ndarrays_o.push_back(out.Reorder2Default());
+      out_blobs.emplace_back(temp_ndarrays_o.back().data());
+    } else {
+      out_blobs.emplace_back(out.data());
+    }
+  }
+
+  std::vector<TBlob> in_data(in_blobs.begin(), in_blobs.begin() + 3);
+  std::vector<TBlob> out_data{in_blobs[3]};
+  std::vector<TBlob> out_grad{in_blobs[4]};
+  const std::vector<TBlob> &in_grad = out_blobs;
+
+  int dtype = in_blobs[rnn_enum::kData].type_flag_;
+  MSHADOW_REAL_TYPE_SWITCH(dtype, DType, {
+    RNNOp<cpu, DType>& op = state_ptr.get_state<RNNOp<cpu, DType>>();
+    const RNNParam& param = op.param_;
+    int index = 5;
+    if (param.state_outputs) {
+      out_data.push_back(in_blobs[index++]);
+      out_grad.push_back(in_blobs[index++]);
+    }
+
+    if (param.mode == rnn_enum::kLstm) {
+      in_data.push_back(in_blobs[index++]);
+      if (param.state_outputs) {
+        out_data.push_back(in_blobs[index++]);
+        out_grad.push_back(in_blobs[index]);
+      }
+    }
+
+    op.Backward(ctx, out_grad, in_data, out_data, req, in_grad);
+  });
+}
+
 #endif
 /*
 index description
diff --git a/src/operator/rnn.cc b/src/operator/rnn.cc
index 4d66becbeba7..da81107ae727 100644
--- a/src/operator/rnn.cc
+++ b/src/operator/rnn.cc
@@ -258,11 +258,9 @@ The definition of GRU here is slightly different from paper but compatible with
 .set_attr<FInferStorageType>("FInferStorageType", RNNStorageType)
 .set_attr<FCreateOpState>("FCreateOpState", CreateRNNState)
 .set_attr<FStatefulCompute>("FStatefulCompute<cpu>", RNNStatefulCompute<cpu>)
-/*
 #if MXNET_USE_MKLDNN == 1
 .set_attr<FStatefulComputeEx>("FStatefulComputeEx<cpu>", RNNStatefulComputeCPU)
 #endif
-*/
 .set_attr<nnvm::FGradient>("FGradient", RNNGrad{"_backward_RNN"})
 .set_attr<FResourceRequestEx>("FResourceRequestEx",
   [](const NodeAttrs& attrs, const int dev_mask, const DispatchMode dispatch_mode) {
@@ -295,6 +293,9 @@ NNVM_REGISTER_OP(_backward_RNN)
 .set_attr_parser(ParamParser<RNNParam>)
 .set_attr<bool>("TIsLayerOpBackward", true)
 .set_attr<nnvm::TIsBackward>("TIsBackward", true)
+#if MXNET_USE_MKLDNN == 1
+.set_attr<FStatefulComputeEx>("FStatefulComputeEx<cpu>", RNNStatefulGradComputeCPU)
+#endif
 .set_attr<FStatefulCompute>("FStatefulCompute<cpu>", RNNStatefulGradCompute<cpu>);
 }  // namespace op
 }  // namespace mxnet