diff --git a/quafu/algorithms/ansatz.py b/quafu/algorithms/ansatz.py
index fffca15a..add12394 100644
--- a/quafu/algorithms/ansatz.py
+++ b/quafu/algorithms/ansatz.py
@@ -145,7 +145,7 @@ class QuantumNeuralNetwork(Ansatz):
     """A Wrapper of quantum circuit as QNN"""
 
     # TODO(zhaoyilun): docs
-    def __init__(self, num_qubits: int, layers: List[Any], interface="torch", backend="sim"):
+    def __init__(self, num_qubits: int, layers: List[Any], interface="torch", backend="sim", sync: bool = False):
         """"""
         # Get transformer according to specified interface
         self._transformer = InterfaceProvider.get(interface)
@@ -154,6 +154,7 @@ def __init__(self, num_qubits: int, layers: List[Any], interface="torch", backen
         self._weights = None
 
         self._backend = backend
+        self._sync = sync
         super().__init__(num_qubits)
 
     def __call__(self, inputs):
@@ -161,7 +162,7 @@ def __call__(self, inputs):
         # pylint: disable=import-outside-toplevel
         from quafu.algorithms.estimator import Estimator
 
-        estimator = Estimator(self, backend=self._backend)
+        estimator = Estimator(self, backend=self._backend, sync=self._sync)
         return self._transformer.execute(self, inputs, estimator=estimator)
 
     def _build(self):
diff --git a/quafu/algorithms/estimator.py b/quafu/algorithms/estimator.py
index 6033bb2c..642e1ce0 100644
--- a/quafu/algorithms/estimator.py
+++ b/quafu/algorithms/estimator.py
@@ -37,17 +37,26 @@ def execute_circuit(circ: QuantumCircuit, observables: Hamiltonian):
 class Estimator:
     """Estimate expectation for quantum circuits and observables"""
 
-    def __init__(self, circ: QuantumCircuit, backend: str = "sim", task: Optional[Task] = None, **task_options) -> None:
+    def __init__(
+        self,
+        circ: QuantumCircuit,
+        backend: str = "sim",
+        sync: bool = False,
+        task: Optional[Task] = None,
+        **task_options,
+    ) -> None:
         """
         Args:
             circ: quantum circuit.
             backend: run on simulator (sim) or real machines (ScQ-PXX)
+            sync: specify whether to send tasks to real machines synchronously
             task: task instance for real machine execution (should be none if backend is "sim")
             task_options: options to config a task instance
         """
         self._circ = circ
         self._circ.get_parameter_grads()  # parameter shift currently requires calling this for initialization
         self._backend = backend
+        self._sync = sync
         self._task = None
         if backend != "sim":
             if task is not None:
@@ -113,16 +122,22 @@ def _run_real_machine(self, observables: Hamiltonian, cache_key: Optional[str] =
             for measure_base in measure_basis:
                 res = self._measure_obs(self._circ, measure_base=measure_base)
                 self._circ.gates = copy.deepcopy(inputs)
-                lst_task_id.append(res.taskid)
-
-            for tid in lst_task_id:
-                # retrieve task results
-                while True:
-                    res = self._task.retrieve(tid)
-                    if res.task_status == "Completed":
-                        exec_res.append(res)
-                        break
-                    time.sleep(0.2)
+                if self._sync:
+                    # directly add res to exec_res
+                    exec_res.append(res)
+                else:
+                    # record task id and retrieve later
+                    lst_task_id.append(res.taskid)
+
+            if not self._sync:
+                for tid in lst_task_id:
+                    # retrieve task results
+                    while True:
+                        res = self._task.retrieve(tid)
+                        if res.task_status == "Completed":
+                            exec_res.append(res)
+                            break
+                        time.sleep(0.2)
 
             if cache_key is not None:
                 # put into cache
@@ -156,7 +171,7 @@ def _measure_obs(self, qc: QuantumCircuit, measure_base: Optional[List] = None)
                 elif base == "Y":
                     qc.rx(pos, np.pi / 2)
 
-            res = self._task.send(qc)
+            res = self._task.send(qc, wait=self._sync)
             res.measure_base = measure_base
 
         return res
diff --git a/quafu/algorithms/gradients/vjp.py b/quafu/algorithms/gradients/vjp.py
index a089ac29..c714a87b 100644
--- a/quafu/algorithms/gradients/vjp.py
+++ b/quafu/algorithms/gradients/vjp.py
@@ -36,6 +36,7 @@ def run_circ(
     circ: QuantumCircuit,
     params: Optional[List[float]] = None,
     backend: str = "sim",
+    sync: bool = False,
     estimator: Optional[Estimator] = None,
 ):
     """Execute a circuit
@@ -44,11 +45,12 @@ def run_circ(
         circ (QuantumCircuit): circ
         params (Optional[List[float]]): params
         backend (str): backend
+        sync (bool): specify whether to submit tasks synchronously when using real machines
         estimator (Optional[Estimator]): estimator
     """
     obs_list = _generate_expval_z(circ.num)
     if estimator is None:
-        estimator = Estimator(circ, backend=backend)
+        estimator = Estimator(circ, backend=backend, sync=sync)
     if params is None:
         params = [g.paras for g in circ.parameterized_gates]
     output = [estimator.run(obs, params, cache_key="00") for obs in obs_list]