diff --git a/requirements.test.txt b/requirements.test.txt
index 29787e2db0..a67a03da9f 100644
--- a/requirements.test.txt
+++ b/requirements.test.txt
@@ -17,4 +17,6 @@ pytest-xdist
 pyyaml
 pythonnet==3.0.1
 clr_loader==0.2.4
-toml==0.10.2
\ No newline at end of file
+toml==0.10.2
+pandas
+tabulate
\ No newline at end of file
diff --git a/tests/compare_util.py b/tests/compare_util.py
index 6398bf65b4..9395cbca21 100644
--- a/tests/compare_util.py
+++ b/tests/compare_util.py
@@ -95,7 +95,7 @@ def compare_ndarray(expected: np.ndarray,
     if dump_hist:
         y, x = np.histogram(expected - actual, 100)
         np.savetxt(dump_file, np.stack((x[:-1], y)).T, fmt='%f', delimiter=',')
-    similarity_info = f"\n{similarity_name} similarity = {similarity}, threshold = {threshold}\n"
+    similarity_info = f"{similarity_name} similarity = {similarity}, threshold = {threshold}"
 
     if similarity_name in ['cosine', 'euclidean', 'segment']:
         compare_op = lt
diff --git a/tests/config.toml b/tests/config.toml
index a743bb958a..93825dc482 100644
--- a/tests/config.toml
+++ b/tests/config.toml
@@ -1,6 +1,7 @@
 name = 'default_config'
 root = 'tests_output'
 dump_hist = false
+dump_infer = false
 
 [compile_opt]
 preprocess = false
diff --git a/tests/inference.py b/tests/inference.py
index c8052efa30..3b6b696b8a 100644
--- a/tests/inference.py
+++ b/tests/inference.py
@@ -7,6 +7,7 @@
 import socket
 import json
 from test_utils import *
+import time
 
 
 class Inference:
@@ -19,8 +20,17 @@ def run_inference(self, compiler, target, ptq_enabled, infer_dir):
         running_on_evb = in_ci and target in kpu_targets and nuc_ip is not None and nuc_port is not None and test_executable is not None and len(
             self.inputs) > 0 and len(self.outputs) > 0
 
+        if self.cfg['dump_infer']:
+            self.infer_dict['case'] = os.path.basename(self.case_dir)
+            self.infer_dict['target'] = target
         if ptq_enabled:
             self.set_quant_opt(compiler)
+
+            if self.cfg['dump_infer']:
+                case = os.path.basename(self.case_dir)
+                self.infer_dict['if_quant_type'] = self.cfg['ptq_opt']['quant_type']
+                self.infer_dict['w_quant_type'] = self.cfg['ptq_opt']['w_quant_type']
+
         compiler.compile()
         kmodel = compiler.gencode_tobytes()
         os.makedirs(infer_dir, exist_ok=True)
@@ -35,7 +45,17 @@ def run_inference(self, compiler, target, ptq_enabled, infer_dir):
             sim = nncase.Simulator()
             sim.load_model(kmodel)
             self.set_infer_input(sim, compile_opt)
+
+            if self.cfg['dump_infer']:
+                t1 = time.perf_counter()
+
             sim.run()
+
+            if self.cfg['dump_infer']:
+                t = (time.perf_counter() - t1) * 1000
+                self.infer_dict['time(ms)'] = str(t)
+                self.infer_dict['fps'] = str(round(1000 / t, 2))
+
             outputs = self.dump_infer_output(sim, compile_opt, infer_dir)
         return outputs
 
@@ -126,8 +146,15 @@ def run_evb(self, target, kmodel, compile_opt):
 
         # get infer result
         outputs = []
-        cmd_result = client_socket.recv(1024).decode()
-        if cmd_result.find('finish') != -1:
+        result_dict = {}
+        ret = client_socket.recv(1024)
+        result_dict = json.loads(ret.decode())
+        if result_dict['type'].find('finish') != -1:
+            if self.cfg['dump_infer']:
+                t = result_dict['time']
+                self.infer_dict['time(ms)'] = str(t)
+                self.infer_dict['fps'] = str(round(1000 / t, 2))
+
             client_socket.sendall(f"pls send outputs".encode())
 
             # recv outputs
@@ -150,6 +177,11 @@ def run_evb(self, target, kmodel, compile_opt):
             client_socket.close()
         else:
             client_socket.close()
-            raise Exception(f'{cmd_result}')
+
+            if self.cfg['dump_infer']:
+                self.infer_dict['result'] = 'Fail'
+                self.infer_dict['remark'] = result_dict['error']
+                dump_dict_to_json(self.infer_dict, self.infer_file)
+            raise Exception(result_dict['error'])
 
         return outputs
diff --git a/tests/json2md.py b/tests/json2md.py
new file mode 100644
index 0000000000..983d859166
--- /dev/null
+++ b/tests/json2md.py
@@ -0,0 +1,24 @@
+import argparse
+import json
+import pandas as pd
+
+
+def json2md(json_file):
+    json_list = []
+    with open(json_file, 'r') as f:
+        json_list = json.load(f)
+
+    json_list = sorted(json_list, key=lambda d: d['case'])
+    df = pd.DataFrame.from_records(json_list)
+    md = df.to_markdown()
+    md_file = json_file.split('/')[-1].split('.')[0] + '.md'
+
+    with open(md_file, 'w') as f:
+        f.write(md)
+
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(prog="json2md")
+    parser.add_argument("--json", help='json file', type=str)
+    args = parser.parse_args()
+    json2md(args.json)
diff --git a/tests/nuc_proxy.py b/tests/nuc_proxy.py
index bc9dcab969..f2ffdac1fa 100644
--- a/tests/nuc_proxy.py
+++ b/tests/nuc_proxy.py
@@ -19,7 +19,7 @@ def __init__(self, port, baudrate, logger):
         self.port = port
         self.baudrate = baudrate
         self.logger = logger
-        self.timeout = 20
+        self.timeout = 60
 
     def open(self):
         self.logger.debug(f'open {self.port} begin')
@@ -144,17 +144,21 @@ def infer_worker(target):
 
         for cmd in cmds.split(';'):
             ret = target.s1.run_cmd(cmd, separator)
-            target.logger.debug("ret = {0}".format(ret))
 
         # infer result
+        dict = {'type': 'finish', 'time': 0.0, 'error': ''}
         if ret.find('terminate') != -1 or ret.find('Exception') != -1:
-            err = f'infer exception: {ret}'
             target.logger.error('infer exception')
-            conn.sendall(err[0:1024].encode())
+            err = f'infer exception: {ret}'
+            dict['type'] = 'exception'
+            dict['error'] = err[0:1024]
+            conn.sendall(json.dumps(dict).encode())
         elif ret.find(separator) == -1:
             # reboot target when timeout
-            conn.sendall(f'infer timeout'.encode())
-            target.logger.error('reboot {0} for timeout'.format(target.name))
+            target.logger.error('reboot for timeout')
+            dict['type'] = 'timeout'
+            dict['error'] = 'infer timeout'
+            conn.sendall(json.dumps(dict).encode())
 
             # reboot after login
             target.s0.run_cmd('root')
@@ -162,7 +166,8 @@ def infer_worker(target):
             target.s0.run_cmd('reboot')
             time.sleep(20)
         else:
-            conn.sendall(f'infer finish'.encode())
+            dict['time'] = float(ret.split('\n')[1].split()[1])
+            conn.sendall(json.dumps(dict).encode())
             dummy = conn.recv(1024)
 
             # send outputs
diff --git a/tests/test_runner.py b/tests/test_runner.py
index 9d8c03b28f..2e0b89055e 100644
--- a/tests/test_runner.py
+++ b/tests/test_runner.py
@@ -55,6 +55,19 @@ def __init__(self, case_name, override_cfg: str = None) -> None:
         # used for tag dynamic model for onnx simplify
         self.dynamic = False
 
+        if self.cfg['dump_infer']:
+            self.infer_file = test_utils.infer_file()
+            self.infer_dict = {
+                'case': 'unknown',
+                'target': 'cpu',
+                'if_quant_type': 'uint8',
+                'w_quant_type': 'uint8',
+                'time(ms)': 'N/A',
+                'fps': 'N/A',
+                'result': 'Pass',
+                'remark': 'N/A'
+            }
+
     def transform_input(self, values: List[np.ndarray], type: str, stage: str) -> List[np.ndarray]:
         new_values = []
         compile_opt = self.cfg['compile_opt']
@@ -252,6 +265,10 @@ def run(self, model_file: Union[List[str], str]):
                             judge, result = self.compare_results(
                                 expected, actual, stage, k_target, v_target['similarity_name'], k_mode, v_mode['threshold'], dump_hist, mode_dir)
 
+                            if stage == 'infer' and self.cfg['dump_infer']:
+                                self.infer_dict['result'] = 'Pass' if judge else 'Fail'
+                                self.infer_dict['remark'] = result.replace('\n', ' ')
+                                dump_dict_to_json(self.infer_dict, self.infer_file)
                             if not judge:
                                 if test_utils.in_ci():
                                     self.clear(self.case_dir)
@@ -407,17 +424,19 @@ def compare_results(self,
                         stage, target, similarity_name, mode, threshold, dump_hist, dump_dir) -> Tuple[bool, str]:
         i = 0
         judges = []
+        result = ''
         for expected, actual in zip(ref_ouputs, test_outputs):
             expected = expected.astype(np.float32)
             actual = actual.astype(np.float32)
             dump_file = os.path.join(dump_dir, 'nncase_result_{0}_hist.csv'.format(i))
             judge, similarity_info = compare_ndarray(
                 expected, actual, similarity_name, threshold, dump_hist, dump_file)
-            result_info = "\n{0} [ {1} {2} {3} ] Output: {4}!!\n".format(
+            result_info = "{0} [ {1} {2} {3} ] Output {4}:".format(
                 'Pass' if judge else 'Fail', stage, target, mode, i)
-            result = similarity_info + result_info
-            with open(os.path.join(self.case_dir, 'test_result.txt'), 'a+') as f:
-                f.write(result)
+            result += result_info + similarity_info
             i = i + 1
             judges.append(judge)
+
+        with open(os.path.join(self.case_dir, 'test_result.txt'), 'a+') as f:
+            f.write(result)
         return sum(judges) == len(judges), result
diff --git a/tests/test_utils.py b/tests/test_utils.py
index b2716c8b9c..3cd04c54bd 100644
--- a/tests/test_utils.py
+++ b/tests/test_utils.py
@@ -1,4 +1,5 @@
 import os
+import json
 import numpy as np
 
 
@@ -33,6 +34,17 @@ def _cast_bfloat16_then_float32(values: np.array):
         values[i] = value
 
 
+def dump_dict_to_json(dict, json_file):
+    json_list = []
+    if os.path.exists(json_file):
+        with open(json_file, 'r') as f:
+            json_list = json.load(f)
+
+    json_list.append(dict)
+    with open(json_file, 'w') as f:
+        json.dump(json_list, f)
+
+
 def in_ci():
     return os.getenv('CI', False)
 
@@ -51,3 +63,7 @@ def nuc_port():
 
 def test_executable(target):
     return os.getenv('TEST_EXECUTABLE_{0}'.format(target.upper()))
+
+
+def infer_file():
+    return os.getenv('INFER_FILE', 'infer_report.json')