diff --git a/python/paddle/fluid/tests/unittests/mkldnn/test_conv2d_mkldnn_op.py b/python/paddle/fluid/tests/unittests/mkldnn/test_conv2d_mkldnn_op.py
index e070f5fbbe057f3290ddb8d2d04d579c98107512..6600d1456d787126203b974db70fde491e26e43f 100644
--- a/python/paddle/fluid/tests/unittests/mkldnn/test_conv2d_mkldnn_op.py
+++ b/python/paddle/fluid/tests/unittests/mkldnn/test_conv2d_mkldnn_op.py
@@ -18,7 +18,7 @@ import unittest
 import numpy as np
 
 import paddle.fluid.core as core
-from paddle.fluid.tests.unittests.op_test import OpTest
+from paddle.fluid.tests.unittests.op_test import OpTest, skip_check_grad_ci
 from paddle.fluid.tests.unittests.test_conv2d_op import TestConv2dOp, TestConv2dOp_v2
 
 
@@ -104,6 +104,8 @@ class TestConv2dMKLDNNOp(TestConv2dOp):
         self.outputs['Output'] = output
 
 
+@skip_check_grad_ci(
+    reason="Fusion is for inference only, check_grad is not required.")
 class TestWithbreluFusion(TestConv2dMKLDNNOp):
     def init_test_case(self):
         TestConv2dMKLDNNOp.init_test_case(self)
@@ -111,16 +113,9 @@ class TestWithbreluFusion(TestConv2dMKLDNNOp):
         self.fuse_alpha = 6.0
         self.dsttype = np.float32
 
-    def test_check_grad(self):
-        pass
-
-    def test_check_grad_no_filter(self):
-        pass
-
-    def test_check_grad_no_input(self):
-        pass
-
 
+@skip_check_grad_ci(
+    reason="Fusion is for inference only, check_grad is not required.")
 class TestWithFuse(TestConv2dMKLDNNOp):
     def init_test_case(self):
         TestConv2dMKLDNNOp.init_test_case(self)
@@ -130,15 +125,6 @@ class TestWithFuse(TestConv2dMKLDNNOp):
         self.fuse_residual_connection = True
         self.input_residual_size = [2, 6, 5, 5]
 
-    def test_check_grad(self):
-        pass
-
-    def test_check_grad_no_filter(self):
-        pass
-
-    def test_check_grad_no_input(self):
-        pass
-
 
 class TestWithPadWithBias(TestConv2dMKLDNNOp):
     def init_test_case(self):
diff --git a/python/paddle/fluid/tests/unittests/mkldnn/test_mul_int8_mkldnn_op.py b/python/paddle/fluid/tests/unittests/mkldnn/test_mul_int8_mkldnn_op.py
index 9280bf886fb6036824bcc1f36c56e43855963f60..0c91868d30245b4dafbaeaf8fba576506758eb30 100644
--- a/python/paddle/fluid/tests/unittests/mkldnn/test_mul_int8_mkldnn_op.py
+++ b/python/paddle/fluid/tests/unittests/mkldnn/test_mul_int8_mkldnn_op.py
@@ -17,12 +17,15 @@ from __future__ import print_function
 import unittest
 import numpy as np
 import paddle.fluid.core as core
-from paddle.fluid.tests.unittests.op_test import OpTest
+from paddle.fluid.tests.unittests.op_test import OpTest, skip_check_grad_ci
 '''
  test case for s8 * s8
 '''
 
 
+@skip_check_grad_ci(
+    reason="mul_mkldnn_op does not implement grad operator, check_grad is not required."
+)
 class TestMKLDNNMulOpS8S8(OpTest):
     def setUp(self):
         self.op_type = "mul"
@@ -77,15 +80,6 @@ class TestMKLDNNMulOpS8S8(OpTest):
         self.check_output_with_place(
             core.CPUPlace(), atol=0, check_dygraph=False)
 
-    def test_check_grad_normal(self):
-        pass
-
-    def test_check_grad_ingore_x(self):
-        pass
-
-    def test_check_grad_ingore_y(self):
-        pass
-
 
 '''
  test case for  s8 * u8 
diff --git a/python/paddle/fluid/tests/unittests/test_conv2d_op.py b/python/paddle/fluid/tests/unittests/test_conv2d_op.py
index 29f45b228024e1ad919e260ef2013d35ed1490f1..8025a332396d6c68b1656955c590cb9ab3243d82 100644
--- a/python/paddle/fluid/tests/unittests/test_conv2d_op.py
+++ b/python/paddle/fluid/tests/unittests/test_conv2d_op.py
@@ -355,7 +355,8 @@ class TestConv2dOp(OpTest):
             place, atol=1e-5, check_dygraph=(self.use_mkldnn == False))
 
     def test_check_grad(self):
-        if self.dtype == np.float16:
+        if self.dtype == np.float16 or (hasattr(self, "no_need_check_grad") and
+                                        self.no_need_check_grad == True):
             return
         place = core.CUDAPlace(0) if self.has_cuda() else core.CPUPlace()
         # TODO(wangzhongpu): support mkldnn op in dygraph mode
@@ -366,7 +367,8 @@ class TestConv2dOp(OpTest):
             check_dygraph=(self.use_mkldnn == False))
 
     def test_check_grad_no_filter(self):
-        if self.dtype == np.float16:
+        if self.dtype == np.float16 or (hasattr(self, "no_need_check_grad") and
+                                        self.no_need_check_grad == True):
             return
         place = core.CUDAPlace(0) if self.has_cuda() else core.CPUPlace()
         # TODO(wangzhongpu): support mkldnn op in dygraph mode
@@ -378,7 +380,8 @@ class TestConv2dOp(OpTest):
             check_dygraph=(self.use_mkldnn == False))
 
     def test_check_grad_no_input(self):
-        if self.dtype == np.float16:
+        if self.dtype == np.float16 or (hasattr(self, "no_need_check_grad") and
+                                        self.no_need_check_grad == True):
             return
         place = core.CUDAPlace(0) if self.has_cuda() else core.CPUPlace()
         # TODO(wangzhongpu): support mkldnn op in dygraph mode