diff --git a/onnxruntime/core/optimizer/nchwc_transformer.cc b/onnxruntime/core/optimizer/nchwc_transformer.cc
index 1fc98e3e98..b19f167c54 100644
--- a/onnxruntime/core/optimizer/nchwc_transformer.cc
+++ b/onnxruntime/core/optimizer/nchwc_transformer.cc
@@ -684,7 +684,7 @@ void NchwcTransformerImpl::Transform(Node& node) {
     if (graph_utils::IsSupportedOptypeVersionAndDomain(node, "Add", {7}) ||
         graph_utils::IsSupportedOptypeVersionAndDomain(node, "Sum", {6, 8})) {
       TransformAdd(node);
-    } else if (graph_utils::IsSupportedOptypeVersionAndDomain(node, "Concat", {4})) {
+    } else if (graph_utils::IsSupportedOptypeVersionAndDomain(node, "Concat", {4, 11})) {
       TransformConcat(node);
     } else if (graph_utils::IsSupportedOptypeVersionAndDomain(node, "Relu", {6})) {
       TransformActivation(node);
diff --git a/onnxruntime/test/optimizer/nchwc_optimizer_test.cc b/onnxruntime/test/optimizer/nchwc_optimizer_test.cc
index 69f3e8be1a..0fd8746075 100644
--- a/onnxruntime/test/optimizer/nchwc_optimizer_test.cc
+++ b/onnxruntime/test/optimizer/nchwc_optimizer_test.cc
@@ -79,27 +79,29 @@ struct NchwcTestHelper {
     return &graph_.GetOrCreateNodeArg(name, nullptr);
   }
 
-  NodeArg* MakeInitializer(const std::vector<int64_t>& shape) {
+  NodeArg* MakeInitializer(const std::vector<int64_t>& shape, const std::vector<float>& data) {
     std::string name = graph_.GenerateNodeArgName("constant");
     ONNX_NAMESPACE::TensorProto tensor_proto;
     tensor_proto.set_name(name);
     tensor_proto.set_data_type(ONNX_NAMESPACE::TensorProto_DataType_FLOAT);
 
-    int64_t num_elements = 1;
     for (auto& dim : shape) {
       tensor_proto.add_dims(dim);
-      num_elements *= dim;
     }
 
-    auto random_data = FillRandomData(static_cast<size_t>(num_elements));
-    tensor_proto.mutable_float_data()->Resize(static_cast<int>(num_elements), 0.0f);
-    memcpy(tensor_proto.mutable_float_data()->mutable_data(), random_data.data(), random_data.size() * sizeof(float));
+    tensor_proto.mutable_float_data()->Resize(static_cast<int>(data.size()), 0.0f);
+    memcpy(tensor_proto.mutable_float_data()->mutable_data(), data.data(), data.size() * sizeof(float));
 
     graph_.AddInitializedTensor(tensor_proto);
 
     return &graph_.GetOrCreateNodeArg(name, nullptr);
   }
 
+  NodeArg* MakeInitializer(const std::vector<int64_t>& shape) {
+    int64_t num_elements = std::accumulate(shape.begin(), shape.end(), int64_t(1), std::multiplies<int64_t>{});
+    return MakeInitializer(shape, FillRandomData(static_cast<size_t>(num_elements)));
+  }
+
   Node& AddNode(const std::string& op_type,
                 const std::vector<NodeArg*>& input_args,
                 const std::vector<NodeArg*>& output_args) {
@@ -120,6 +122,21 @@ struct NchwcTestHelper {
     return AddNode("Conv", input_args, {output_arg});
   }
 
+  Node& AddClipNode(NodeArg* input_arg, NodeArg* output_arg, float min, float max) {
+    int opset_version = graph_.DomainToVersionMap().find(kOnnxDomain)->second;
+    std::vector<NodeArg*> input_args = {input_arg};
+    if (opset_version >= 11) {
+      input_args.push_back(MakeInitializer({1}, {min}));
+      input_args.push_back(MakeInitializer({1}, {max}));
+    }
+    auto& node = AddNode("Clip", input_args, {output_arg});
+    if (opset_version < 11) {
+      node.AddAttribute("min", min);
+      node.AddAttribute("max", max);
+    }
+    return node;
+  }
+
   std::vector<float> FillRandomData(size_t count) {
     constexpr int min_fill_value = -23;
     constexpr int max_fill_value = 23;
@@ -144,7 +161,7 @@ struct NchwcTestHelper {
 
 void NchwcOptimizerTester(const std::function<void(NchwcTestHelper& helper)>& build_test_case,
                           const std::function<void(NchwcInferenceSession& session)>& check_nchwc_graph,
-                          int opset_version = 10) {
+                          int opset_version = 11) {
   // Ignore the test if NCHWc is not supported by the platform.
   if (MlasNchwcGetBlockSize() <= 1) {
     return;
@@ -215,10 +232,10 @@ TEST(NchwcOptimizerTests, ConvNchw) {
       auto* conv_output_arg = output_arg;
       if (!activation_op_type.empty()) {
         conv_output_arg = helper.MakeIntermediate();
-        auto& act_node = helper.AddNode(activation_op_type, {conv_output_arg}, {output_arg});
         if (activation_op_type == "Clip") {
-          act_node.AddAttribute("min", 0.0f);
-          act_node.AddAttribute("max", 6.0f);
+          helper.AddClipNode(conv_output_arg, output_arg, 0.0f, 6.0f);
+        } else {
+          helper.AddNode(activation_op_type, {conv_output_arg}, {output_arg});
         }
       }
 
@@ -255,10 +272,10 @@ TEST(NchwcOptimizerTests, ConvNchwc) {
       auto* conv_output_arg = output_arg;
       if (!activation_op_type.empty()) {
         conv_output_arg = helper.MakeIntermediate();
-        auto& act_node = helper.AddNode(activation_op_type, {conv_output_arg}, {output_arg});
         if (activation_op_type == "Clip") {
-          act_node.AddAttribute("min", -6.0f);
-          act_node.AddAttribute("max", 6.0f);
+          helper.AddClipNode(conv_output_arg, output_arg, -6.0f, 6.0f);
+        } else {
+          helper.AddNode(activation_op_type, {conv_output_arg}, {output_arg});
         }
       }
 
@@ -533,7 +550,7 @@ TEST(NchwcOptimizerTests, ConvAddFusion) {
   // Verify that Add or Sum can be fused into a preceding NCHWc Conv node,
   // with an optional Relu node following.
   std::vector<std::string> op_types = {"Add", "Sum"};
-  static const int opset_versions[] = {7, 10};
+  static const int opset_versions[] = {7, 10, 11};
   for (auto& op_type : op_types) {
     for (auto opset_version : opset_versions) {
       test_case(op_type, opset_version, false);