diff --git a/neural_speed/core/ne_layers.c b/neural_speed/core/ne_layers.c
index a6154bda2..5a275f0d5 100644
--- a/neural_speed/core/ne_layers.c
+++ b/neural_speed/core/ne_layers.c
@@ -10882,6 +10882,7 @@ void ne_graph_compute(struct ne_context* ctx, struct ne_cgraph* cgraph) {
 }
 
 void ne_graph_profiling(const struct ne_cgraph* cgraph) {
+#ifdef NS_PERF
   int64_t perf_total_per_op_us[NE_OP_COUNT] = {0};
 
   NE_PRINT("=== GRAPH Profiling ===\n");
@@ -10904,6 +10905,10 @@ void ne_graph_profiling(const struct ne_cgraph* cgraph) {
   }
   NE_PRINT("perf_total_per_op_us[%24s] = %7.3f ms\n", "INNER PRODUCT", (double)ip_duration / 1000.0);
   NE_PRINT("========================================\n");
+
+#else
+  NE_PRINT("\n[Warning] To collect profiling data, please recompile with NS_PROFILING=ON.\n");
+#endif
 }
 
 void ne_graph_reset(struct ne_cgraph* cgraph) {
diff --git a/setup.py b/setup.py
index fbda29af2..319cb6030 100644
--- a/setup.py
+++ b/setup.py
@@ -93,6 +93,7 @@ def build_extension(self, ext: CMakeExtension) -> None:
         extdir = ext_fullpath.parent.resolve()
 
         output_dir = f"{extdir}{os.sep}"
+        NS_PROFILING_ENV = os.environ.get("NS_PROFILING", "OFF")
         cmake_args = [
             f"-DCMAKE_LIBRARY_OUTPUT_DIRECTORY={output_dir}",
             f"-DCMAKE_RUNTIME_OUTPUT_DIRECTORY={output_dir}",
@@ -104,6 +105,7 @@ def build_extension(self, ext: CMakeExtension) -> None:
             f"-DNS_WITH_AVX2={'ON' if NS_WITH_AVX2 else 'OFF'}",
             f"-DNS_WITH_TESTS=OFF",
             f"-DNS_PYTHON_API=ON",
+            f"-DNS_PROFILING={NS_PROFILING_ENV}",
         ]
         if sys.platform == "linux":  # relative_rpath
             cmake_args.append('-DCMAKE_BUILD_RPATH=$ORIGIN/')