diff --git a/tf2jax/_src/ops_test.py b/tf2jax/_src/ops_test.py
index 2a6da77..aa4e467 100644
--- a/tf2jax/_src/ops_test.py
+++ b/tf2jax/_src/ops_test.py
@@ -339,8 +339,6 @@ def test_avg_pool(self, padding, data_format):
     strides = (1, 3, 2, 1)
 
     if data_format == "NCHW":
-      if jax.default_backend().lower() == "cpu":
-        self.skipTest("TensorFlow AvgPool does not support NCHW on CPU.")
       inputs = np.transpose(inputs, [0, 3, 1, 2])
       ksize = _reorder(ksize, [0, 3, 1, 2])
       strides = _reorder(strides, [0, 3, 1, 2])
@@ -620,8 +618,6 @@ def test_conv2d(self, input_shape, filter_shape, strides, padding,
     filters = np.random.normal(size=filter_shape).astype(np.float32)
     inputs = np.random.normal(size=input_shape).astype(np.float32)
     if data_format == "NCHW":
-      if jax.default_backend().lower() == "cpu":
-        self.skipTest("TensorFlow Conv2D does not support NCHW on CPU.")
       inputs = np.transpose(inputs, [0, 3, 1, 2])
       strides = _reorder(strides, [0, 3, 1, 2])
       dilations = _reorder(dilations, [0, 3, 1, 2])
@@ -664,9 +660,6 @@ def test_conv2d_transpose(self, data_format):
     filters = np.random.normal(size=[7, 7, 128, 13]).astype(np.float32)
     inputs = np.random.normal(size=[3, 4, 4, 13]).astype(np.float32)
     if data_format == "NCHW":
-      if jax.default_backend().lower() == "cpu":
-        self.skipTest(
-            "TensorFlow Conv2DBackpropInput does not support NCHW on CPU.")
       inputs = np.transpose(inputs, [0, 3, 1, 2])
       strides = _reorder(strides, [0, 3, 1, 2])
       dilations = _reorder(dilations, [0, 3, 1, 2])
@@ -768,9 +761,6 @@ def test_depthwise_conv2d(self, use_explicit_paddings, data_format):
     dilations = [1, 1, 1, 1]
 
     if data_format == "NCHW":
-      if jax.default_backend().lower() == "cpu":
-        self.skipTest(
-            "TensorFlow DepthwiseConv2dNative does not support NCHW on CPU.")
       inputs = np.transpose(inputs, [0, 3, 1, 2])
       strides = _reorder(strides, [0, 3, 1, 2])
       dilations = _reorder(dilations, [0, 3, 1, 2])
@@ -1081,13 +1071,6 @@ def test_igamma_ops(self):
 
   @chex.variants(with_jit=True, without_jit=True)
   def test_inplace_add(self):
-    if test_util.parse_version(tf.version.VERSION) >= test_util.parse_version(
-        "2.14.0"
-    ):
-      self.skipTest(
-          f"Requires earlier than tf 2.14.0, found {tf.version.VERSION}."
-      )
-
     np.random.seed(42)
 
     @tf.function
@@ -1116,13 +1099,6 @@ def inplace_add(x, idx, val):
 
   @chex.variants(with_jit=True, without_jit=True)
   def test_inplace_update(self):
-    if test_util.parse_version(tf.version.VERSION) >= test_util.parse_version(
-        "2.14.0"
-    ):
-      self.skipTest(
-          f"Requires earlier than tf 2.14.0, found {tf.version.VERSION}."
-      )
-
     np.random.seed(42)
 
     @tf.function
@@ -1278,8 +1254,6 @@ def test_max_pool(self, padding, data_format):
     strides = (1, 3, 2, 1)
 
     if data_format == "NCHW":
-      if jax.default_backend().lower() == "cpu":
-        self.skipTest("TensorFlow MaxPool does not support NCHW on CPU.")
       inputs = np.transpose(inputs, [0, 3, 1, 2])
       ksize = _reorder(ksize, [0, 3, 1, 2])
       strides = _reorder(strides, [0, 3, 1, 2])
@@ -2186,9 +2160,6 @@ def while_loop(x):
       return tf.while_loop(cond, body, [x, step])
     self._test_convert(while_loop, inputs)
 
-    if jax.default_backend().lower() == "gpu":
-      self.skipTest("Skip remaining tests on GPU due to CUDA errors.")
-
     def raw_stateless_while(x):
       loop_vars = [x, step]
       return tf.raw_ops.StatelessWhile(
diff --git a/tf2jax/_src/roundtrip_test.py b/tf2jax/_src/roundtrip_test.py
index 8738fff..5302710 100644
--- a/tf2jax/_src/roundtrip_test.py
+++ b/tf2jax/_src/roundtrip_test.py
@@ -1166,11 +1166,6 @@ def forward(a, b):
         grad_tols=tols)
 
   def test_explicit_native_serialization(self):
-    if test_util.parse_version(tf.version.VERSION) < test_util.parse_version(
-        "2.12.0"
-    ):
-      self.skipTest(f"Requires tf 2.12.0 or later, found {tf.version.VERSION}.")
-
     def forward(x):
       return x + 3.14