diff --git a/onedal/decomposition/pca.cpp b/onedal/decomposition/pca.cpp
index 49226f3265..11b48c9d8a 100644
--- a/onedal/decomposition/pca.cpp
+++ b/onedal/decomposition/pca.cpp
@@ -112,6 +112,10 @@ void init_train_result(py::module_& m) {
         .def_property_readonly("explained_variances_ratio",
                                &result_t::get_explained_variances_ratio)
 #endif // defined(ONEDAL_VERSION) && ONEDAL_VERSION>=20240100
+#if defined(ONEDAL_VERSION) && ONEDAL_VERSION >= 20250100
+        .def_property_readonly("noise_variance",
+                                &result_t::get_noise_variance)                      
+#endif // defined(ONEDAL_VERSION) && ONEDAL_VERSION>=20250100
         .def_property_readonly("means", &result_t::get_means)
         .def_property_readonly("variances", &result_t::get_variances);
 }
diff --git a/onedal/decomposition/pca.py b/onedal/decomposition/pca.py
index aa251489a8..d46ea9c651 100644
--- a/onedal/decomposition/pca.py
+++ b/onedal/decomposition/pca.py
@@ -104,22 +104,6 @@ def _resolve_n_components_for_result(self, shape_tuple):
         else:
             return self.n_components
 
-    def _compute_noise_variance(self, n_components, n_sf_min):
-        if n_components < n_sf_min:
-            if len(self.explained_variance_) == n_sf_min:
-                return self.explained_variance_[n_components:].mean()
-            elif len(self.explained_variance_) < n_sf_min:
-                # TODO Rename variances_ to var_ to align with sklearn/sklearnex IncrementalPCA
-                if hasattr(self, "variances_"):
-                    resid_var = self.variances_.sum()
-                elif hasattr(self, "var_"):
-                    resid_var = self.var_.sum()
-
-                resid_var -= self.explained_variance_.sum()
-                return resid_var / (n_sf_min - n_components)
-        else:
-            return 0.0
-
     def _create_model(self):
         # Not supported with spmd policy so BasePCA must be specified
         m = BasePCA._get_backend(BasePCA, "decomposition", "dim_reduction", "model")
@@ -185,6 +169,7 @@ def fit(self, X, y=None, queue=None):
         self.explained_variance_ratio_ = from_table(
             result.explained_variances_ratio
         ).ravel()
+        self.noise_variance_ = from_table(result.noise_variance_).ravel()
         self.n_samples_ = n_samples
         self.n_features_ = n_features
 
@@ -194,7 +179,6 @@ def fit(self, X, y=None, queue=None):
 
         n_components = self._resolve_n_components_for_result(X.shape)
         self.n_components_ = n_components
-        self.noise_variance_ = self._compute_noise_variance(n_components, n_sf_min)
 
         if n_components < params["n_components"]:
             self.explained_variance_ = self.explained_variance_[:n_components]
diff --git a/onedal/decomposition/tests/test_incremental_pca.py b/onedal/decomposition/tests/test_incremental_pca.py
index 94a0b1acc6..95701c610e 100644
--- a/onedal/decomposition/tests/test_incremental_pca.py
+++ b/onedal/decomposition/tests/test_incremental_pca.py
@@ -183,8 +183,8 @@ def test_on_random_data(
         if len(sorted_eigenvalues) > n_components
         else 0.0
     )
-    # TODO Fix noise variance computation (It is necessary to update C++ side)
-    # assert np.abs(incpca.noise_variance_ - expected_noise_variance) < tol
+
+    assert np.abs(incpca.noise_variance_ - expected_noise_variance) < tol
 
     expected_transformed_data = centered_data @ components.T
     if whiten:
diff --git a/sklearnex/preview/decomposition/tests/test_incremental_pca.py b/sklearnex/preview/decomposition/tests/test_incremental_pca.py
index c4c47c8adb..f34c6c93c2 100644
--- a/sklearnex/preview/decomposition/tests/test_incremental_pca.py
+++ b/sklearnex/preview/decomposition/tests/test_incremental_pca.py
@@ -167,8 +167,8 @@ def check_pca(incpca, dtype, whiten, data, transformed_data):
         if len(sorted_eigenvalues) > n_components
         else 0.0
     )
-    # TODO Fix noise variance computation (It is necessary to update C++ side)
-    # assert np.abs(incpca.noise_variance_ - expected_noise_variance) < tol
+
+    assert np.abs(incpca.noise_variance_ - expected_noise_variance) < tol
 
     expected_transformed_data = centered_data @ components.T
     if whiten: