diff --git a/01.fit_a_line/train.py b/01.fit_a_line/train.py
index 255180d3c4322e8dd201e96917e288e3ee209d61..d18ab4549f57f294c43c91e67fbceca69115c57f 100644
--- a/01.fit_a_line/train.py
+++ b/01.fit_a_line/train.py
@@ -1,10 +1,12 @@
+import os
 import paddle.v2 as paddle
 import paddle.v2.dataset.uci_housing as uci_housing
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
 
 def main():
     # init
-    paddle.init(use_gpu=False, trainer_count=1)
+    paddle.init(use_gpu=with_gpu, trainer_count=1)
 
     # network config
     x = paddle.layer.data(name='x', type=paddle.data_type.dense_vector(13))
diff --git a/02.recognize_digits/train.py b/02.recognize_digits/train.py
index 52f09b78c781868e55655bc05bf8aa66926304a3..dd0d646f7f62794a04788f2ec76957dd81c9b58e 100644
--- a/02.recognize_digits/train.py
+++ b/02.recognize_digits/train.py
@@ -3,6 +3,7 @@ from PIL import Image
 import numpy as np
 import paddle.v2 as paddle
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
 
 def softmax_regression(img):
     predict = paddle.layer.fc(
@@ -49,7 +50,7 @@ def convolutional_neural_network(img):
 
 
 def main():
-    paddle.init(use_gpu=False, trainer_count=1)
+    paddle.init(use_gpu=with_gpu, trainer_count=1)
 
     # define network topology
     images = paddle.layer.data(
diff --git a/03.image_classification/train.py b/03.image_classification/train.py
index 0c800308ed2b19a86e27b5114b53a8b69eb7a77c..567dbbf1ad2158de1434be0911bb2bc8b6aca4a0 100644
--- a/03.image_classification/train.py
+++ b/03.image_classification/train.py
@@ -12,20 +12,21 @@
 # See the License for the specific language governing permissions and
 # limitations under the License
 
-import sys
+import sys, os
 
 import paddle.v2 as paddle
 
 from vgg import vgg_bn_drop
 from resnet import resnet_cifar10
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
 
 def main():
     datadim = 3 * 32 * 32
     classdim = 10
 
     # PaddlePaddle init
-    paddle.init(use_gpu=False, trainer_count=1)
+    paddle.init(use_gpu=with_gpu, trainer_count=1)
 
     image = paddle.layer.data(
         name="image", type=paddle.data_type.dense_vector(datadim))
diff --git a/04.word2vec/train.py b/04.word2vec/train.py
index eb596673ce8ad55dfe8c5c258beb344e825b7c25..550ba0093d62f50b674eef3663f13323f57dbd3c 100644
--- a/04.word2vec/train.py
+++ b/04.word2vec/train.py
@@ -1,7 +1,9 @@
-import math
+import math, os
 
 import paddle.v2 as paddle
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
+
 embsize = 32
 hiddensize = 256
 N = 5
@@ -17,7 +19,7 @@ def wordemb(inlayer):
 
 
 def main():
-    paddle.init(use_gpu=False, trainer_count=3)
+    paddle.init(use_gpu=with_gpu, trainer_count=3)
     word_dict = paddle.dataset.imikolov.build_dict()
     dict_size = len(word_dict)
     # Every layer takes integer value of range [0, dict_size)
diff --git a/05.recommender_system/train.py b/05.recommender_system/train.py
index cb549e49abca240ea3268bc443aa37568c724cbf..cfe234dad7bd1fb7cb22de889e3a235235781b49 100644
--- a/05.recommender_system/train.py
+++ b/05.recommender_system/train.py
@@ -1,7 +1,9 @@
 import paddle.v2 as paddle
 import cPickle
 import copy
+import os
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
 
 def get_usr_combined_features():
     uid = paddle.layer.data(
@@ -67,7 +69,7 @@ def get_mov_combined_features():
 
 
 def main():
-    paddle.init(use_gpu=False)
+    paddle.init(use_gpu=with_gpu)
     usr_combined_features = get_usr_combined_features()
     mov_combined_features = get_mov_combined_features()
     inference = paddle.layer.cos_sim(
diff --git a/06.understand_sentiment/train.py b/06.understand_sentiment/train.py
index 7878f00b6401ed0e6a0863d2cec129b6e51b163d..8e967c81ccbd5e9f31235ba8a58e0bf11bda6b98 100644
--- a/06.understand_sentiment/train.py
+++ b/06.understand_sentiment/train.py
@@ -12,9 +12,10 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-import sys
+import sys, os
 import paddle.v2 as paddle
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
 
 def convolution_net(input_dim, class_dim=2, emb_dim=128, hid_dim=128):
     data = paddle.layer.data("word",
@@ -102,7 +103,7 @@ def stacked_lstm_net(input_dim,
 
 if __name__ == '__main__':
     # init
-    paddle.init(use_gpu=False)
+    paddle.init(use_gpu=with_gpu)
 
     #data
     print 'load dictionary...'
diff --git a/07.label_semantic_roles/train.py b/07.label_semantic_roles/train.py
index 94d751dffa5edd620793e8818f19e98477212cdc..d8a3698fcc0c56258de0d9bf231bbd6d63a4315f 100644
--- a/07.label_semantic_roles/train.py
+++ b/07.label_semantic_roles/train.py
@@ -1,9 +1,11 @@
-import math
+import math, os
 import numpy as np
 import paddle.v2 as paddle
 import paddle.v2.dataset.conll05 as conll05
 import paddle.v2.evaluator as evaluator
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
+
 word_dict, verb_dict, label_dict = conll05.get_dict()
 word_dict_len = len(word_dict)
 label_dict_len = len(label_dict)
@@ -118,7 +120,7 @@ def load_parameter(file_name, h, w):
 
 
 def main():
-    paddle.init(use_gpu=False, trainer_count=1)
+    paddle.init(use_gpu=with_gpu, trainer_count=1)
 
     # define network topology
     feature_out = db_lstm()
diff --git a/08.machine_translation/train.py b/08.machine_translation/train.py
index 6f857b0cad9c23bee62de3ad85770d8d1cf538a9..54c74924a859a5fcefb5c2590f584b454e101eb1 100644
--- a/08.machine_translation/train.py
+++ b/08.machine_translation/train.py
@@ -1,8 +1,8 @@
-import sys
+import sys, os
 import numpy as np
-
 import paddle.v2 as paddle
 
+with_gpu = os.getenv('WITH_GPU', '0') != '0'
 
 def save_model(parameters, save_path):
     with open(save_path, 'w') as f:
@@ -135,7 +135,7 @@ def seq_to_seq_net(source_dict_dim,
 
 
 def main():
-    paddle.init(use_gpu=False, trainer_count=1)
+    paddle.init(use_gpu=with_gpu, trainer_count=1)
     is_generating = False
 
     # source and target dict dim.