diff --git a/helion/autotuner/base_search.py b/helion/autotuner/base_search.py
index a1b9f0b00..74ba4db7f 100644
--- a/helion/autotuner/base_search.py
+++ b/helion/autotuner/base_search.py
@@ -115,9 +115,19 @@ def _compute_baseline(self) -> tuple[object, bool, Sequence[object] | None]:
         """
         new_args = self._clone_args(self._original_args)
         baseline_config = self.config_spec.default_config()
-        baseline_output = self.kernel.compile_config(
-            baseline_config, allow_print=False
-        )(*new_args)
+        try:
+            baseline_output = self.kernel.compile_config(
+                baseline_config, allow_print=False
+            )(*new_args)
+            torch.cuda.synchronize()
+        except Exception as e:
+            decorator = self.kernel.format_kernel_decorator(
+                baseline_config, self.settings
+            )
+            raise exc.InvalidConfig(
+                "Default config failed while computing baseline.\n"
+                f"Default config: {decorator}\n"
+            ) from e
         original_args_flat, _ = tree_flatten(self._original_args)
         new_args_flat, _ = tree_flatten(new_args)
         mutated = False
diff --git a/test/test_atomic_ops.py b/test/test_atomic_ops.py
index 349108bf7..eb4828b3f 100644
--- a/test/test_atomic_ops.py
+++ b/test/test_atomic_ops.py
@@ -278,7 +278,7 @@ def test_atomic_add_w_tile_attr(self):
 
     @skipIfRefEager("Error only raises in normal mode")
     def test_atomic_add_device_tensor_error(self):
-        @helion.kernel(static_shapes=True)
+        @helion.kernel(static_shapes=True, use_default_config=True)
         def kernel(x: torch.Tensor) -> torch.Tensor:
             for tile in hl.tile(x.size(0), block_size=128):
                 device_tensor = hl.zeros([tile], dtype=x.dtype)