diff --git a/python/paddle/fluid/tests/unittests/xpu/CMakeLists.txt b/python/paddle/fluid/tests/unittests/xpu/CMakeLists.txt
index c6aaf363138d4..cf70f63580b99 100644
--- a/python/paddle/fluid/tests/unittests/xpu/CMakeLists.txt
+++ b/python/paddle/fluid/tests/unittests/xpu/CMakeLists.txt
@@ -24,5 +24,7 @@ foreach(TEST_OP ${DIST_TEST_OPS})
   py_test_modules(${TEST_OP} MODULES ${TEST_OP})
 endforeach()
 
-set_tests_properties(test_mul_op_xpu PROPERTIES TIMEOUT 120)
 set_tests_properties(test_conv2d_op_xpu PROPERTIES TIMEOUT 120)
+set_tests_properties(test_mul_op_xpu PROPERTIES TIMEOUT 120)
+set_tests_properties(test_matmul_v2_op_xpu PROPERTIES TIMEOUT 900)
+set_tests_properties(test_matmul_op_xpu PROPERTIES TIMEOUT 300)
diff --git a/python/paddle/fluid/tests/unittests/xpu/get_test_cover_info.py b/python/paddle/fluid/tests/unittests/xpu/get_test_cover_info.py
index 3da9e32b015ed..f58c0d4cf074c 100644
--- a/python/paddle/fluid/tests/unittests/xpu/get_test_cover_info.py
+++ b/python/paddle/fluid/tests/unittests/xpu/get_test_cover_info.py
@@ -87,7 +87,6 @@
 xpu_test_op_type_white_list = [
     'dropout_float16',
     'dropout_grad_float16',
-    'matmul_v2_float16',
     "grad_add_float32"  # no api for grad_add, skip
 ]
 xpu_test_device_op_white_list = []
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_matmul_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_matmul_op_xpu.py
index 1c68f8fb6bf16..73f61c2d9d5ba 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_matmul_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_matmul_op_xpu.py
@@ -294,6 +294,10 @@ def setUp(self):
         self.op_type = "matmul"
         self.dtype = np.float32 if not hasattr(self,
                                                'in_type') else self.in_type
+
+        self.__class__.no_need_check_grad = False if not hasattr(
+            self, 'no_need_check_grad') else self.no_need_check_grad
+
         shape_X = [4, 5] if not hasattr(self, 'shape_X') else self.shape_X
         shape_Y = [5, 6] if not hasattr(self, 'shape_Y') else self.shape_Y
         transpose_X = False if not hasattr(self,
@@ -314,12 +318,20 @@ def test_check_output(self):
         self.check_output_with_place(place, atol=1e-3)
 
     def test_check_grad_normal(self):
+        if hasattr(self.__class__, "no_need_check_grad"
+                   ) and self.__class__.no_need_check_grad == True:
+            return
+
         place = paddle.XPUPlace(0)
         self.check_grad_with_place(place, ['X', 'Y'],
                                    'Out',
                                    max_relative_error=5e-2)
 
     def test_check_grad_ignore_x(self):
+        if hasattr(self.__class__, "no_need_check_grad"
+                   ) and self.__class__.no_need_check_grad == True:
+            return
+
         place = paddle.XPUPlace(0)
         self.check_grad_with_place(place, ['Y'],
                                    'Out',
@@ -327,6 +339,10 @@ def test_check_grad_ignore_x(self):
                                    no_grad_set=set("X"))
 
     def test_check_grad_ignore_y(self):
+        if hasattr(self.__class__, "no_need_check_grad"
+                   ) and self.__class__.no_need_check_grad == True:
+            return
+
         place = paddle.XPUPlace(0)
         self.check_grad_with_place(place, ['X'],
                                    'Out',
@@ -351,6 +367,9 @@ def dynamic_create_class(self):
             for transose_x in [True, False]:
                 for transose_y in [True, False]:
                     for batch in batch_size:
+                        no_need_check_grad = False
+                        if batch >= 5:
+                            no_need_check_grad = True
                         class_name = (
                             'TestMatMulOp_dimX_{}_dim_Y_{}_transX_{}_transY_{}_batch_{}'
                             .format(dim_X, dim_Y, transose_x, transose_y,
@@ -362,6 +381,7 @@ def dynamic_create_class(self):
                             'shape_Y': shape_y,
                             'transpose_X': transose_x,
                             'transpose_Y': transose_y,
+                            'no_need_check_grad': no_need_check_grad,
                             'op_type': "matmul"
                         }
                         classes.append([class_name, attr_dict])
diff --git a/python/paddle/fluid/tests/unittests/xpu/test_matmul_v2_op_xpu.py b/python/paddle/fluid/tests/unittests/xpu/test_matmul_v2_op_xpu.py
index 8f31981355403..92b9ae3ae8998 100644
--- a/python/paddle/fluid/tests/unittests/xpu/test_matmul_v2_op_xpu.py
+++ b/python/paddle/fluid/tests/unittests/xpu/test_matmul_v2_op_xpu.py
@@ -80,6 +80,8 @@ def setUp(self):
             self.dtype = self.in_type
             self.config()
             self.op_type = "matmul_v2"
+            if self.dtype == np.float16 or self.dtype == "float16":
+                self.__class__.no_need_check_grad = True
             x = np.random.random(self.x_shape).astype(self.dtype)
             y = np.random.random(self.y_shape).astype(self.dtype)
             # -0.1 ~ 0.1
@@ -99,6 +101,9 @@ def test_check_output(self):
             self.check_output_with_place(place)
 
         def test_check_grad(self):
+            if hasattr(self.__class__, "no_need_check_grad"
+                       ) and self.__class__.no_need_check_grad == True:
+                return
             place = paddle.XPUPlace(0)
             self.check_grad_with_place(place, ['X', 'Y'], 'Out')