diff --git a/python/paddle_serving_server/__init__.py b/python/paddle_serving_server/__init__.py
index d893861e2b0cc1615701191a66aa6ff0bcb53305..08f3febc14ecd0d3c3cde7dc54f17bd143f9d5b3 100644
--- a/python/paddle_serving_server/__init__.py
+++ b/python/paddle_serving_server/__init__.py
@@ -211,6 +211,7 @@ class Server(object):
         tar_name = floder_name + ".tar.gz"
         bin_url = "https://paddle-serving.bj.bcebos.com/bin/" + tar_name
         self.server_path = os.path.join(self.module_path, floder_name)
+
         if not os.path.exists(self.server_path):
             print('Frist time run, downloading PaddleServing components ...')
             r = os.system('wget ' + bin_url + ' --no-check-certificate')
diff --git a/python/paddle_serving_server_gpu/__init__.py b/python/paddle_serving_server_gpu/__init__.py
index 3dd330b18921c81cf17601ff7e52d860f0322f95..abe93a1ce74fa846774425827b9a3ebbdc670450 100644
--- a/python/paddle_serving_server_gpu/__init__.py
+++ b/python/paddle_serving_server_gpu/__init__.py
@@ -20,6 +20,7 @@ import google.protobuf.text_format
 import tarfile
 import socket
 import paddle_serving_server_gpu as paddle_serving_server
+import time
 from .version import serving_server_version
 from contextlib import closing
 import argparse
@@ -215,11 +216,21 @@ class Server(object):
         os.chdir(self.module_path)
         need_download = False
         device_version = "serving-gpu-"
-        floder_name = device_version + serving_server_version
-        tar_name = floder_name + ".tar.gz"
+        folder_name = device_version + serving_server_version
+        tar_name = folder_name + ".tar.gz"
         bin_url = "https://paddle-serving.bj.bcebos.com/bin/" + tar_name
-        self.server_path = os.path.join(self.module_path, floder_name)
+        self.server_path = os.path.join(self.module_path, folder_name)
+
+        download_flag = "{}/{}.is_download".format(self.module_path,
+                                                   folder_name)
+        if os.path.exists(download_flag):
+            os.chdir(self.cur_path)
+            self.bin_path = self.server_path + "/serving"
+            return
+
         if not os.path.exists(self.server_path):
+            os.system("touch {}/{}.is_download".format(self.module_path,
+                                                       folder_name))
             print('Frist time run, downloading PaddleServing components ...')
             r = os.system('wget ' + bin_url + ' --no-check-certificate')
             if r != 0:
@@ -318,4 +329,8 @@ class Server(object):
                       self.gpuid,)
         print("Going to Run Comand")
         print(command)
+        # wait for other process to download server bin
+        while not os.path.exists(self.server_path):
+            time.sleep(1)
+
         os.system(command)
diff --git a/python/paddle_serving_server_gpu/web_service.py b/python/paddle_serving_server_gpu/web_service.py
index db0db25d749728dc99dc1e65c278741d4b4bd5ae..5d507c9475047d6c7eb65a2b2c5799221cf194b5 100755
--- a/python/paddle_serving_server_gpu/web_service.py
+++ b/python/paddle_serving_server_gpu/web_service.py
@@ -66,6 +66,12 @@ class WebService(object):
         return server
 
     def _launch_rpc_service(self, service_idx):
+        if service_idx == 0:
+            self.rpc_service_list[service_idx].check_local_bin()
+            if not self.rpc_service_list[service_idx].use_local_bin:
+                self.rpc_service_list[service_idx].download_bin()
+        else:
+            time.sleep(3)
         self.rpc_service_list[service_idx].run_server()
 
     def prepare_server(self, workdir="", port=9393, device="gpu", gpuid=0):
@@ -160,15 +166,6 @@ class WebService(object):
             if not isinstance(result, dict) and result == -1:
                 result = {"result": "Request Value Error"}
             return result
-            '''
-            feed, fetch = self.preprocess(request.json, request.json["fetch"])
-            if "fetch" in feed:
-                del feed["fetch"]
-            fetch_map = client.predict(feed=feed, fetch=fetch)
-            fetch_map = self.postprocess(
-                feed=request.json, fetch=fetch, fetch_map=fetch_map)
-            return fetch_map
-            '''
 
         app_instance.run(host="0.0.0.0",
                          port=self.port,