diff --git a/paddle/scripts/submit_local.sh.in b/paddle/scripts/submit_local.sh.in
index 8a352b0078d701f797f7202c85bd0e08201ac9b8..bb47ad614ed85923ce5d9704760ec6c5b5ae59ee 100755
--- a/paddle/scripts/submit_local.sh.in
+++ b/paddle/scripts/submit_local.sh.in
@@ -92,6 +92,9 @@ function threads_config() {
     if [ -z "$OPENBLAS_NUM_THREADS" ]; then
       export OPENBLAS_NUM_THREADS=$threads
     fi
+    if [ $threads -gt 1 ] && [ -z "$OPENBLAS_MAIN_FREE" ]; then
+      export OPENBLAS_MAIN_FREE=1
+    fi
   fi
   
 }
diff --git a/python/paddle/v2/__init__.py b/python/paddle/v2/__init__.py
index 0de417df2cb942ce46ff8ac3acc61ae4999ed634..df710c33d0c0ca16d358dac1eb42327e9cd4c7ae 100644
--- a/python/paddle/v2/__init__.py
+++ b/python/paddle/v2/__init__.py
@@ -62,12 +62,15 @@ __all__ = [
 cp.begin_parse()
 
 
-def set_omp_mkl_env_vars(trainer_count):
+def set_env_vars(trainer_count):
     '''Auto set CPU environment if have not set before.
-       export KMP_AFFINITY, OMP_DYNAMIC according to the Hyper Threading status.
-       export OMP_NUM_THREADS, MKL_NUM_THREADS according to trainer_count.
+       For MKL:
+         export KMP_AFFINITY, OMP_DYNAMIC according to the Hyper Threading status.
+         export OMP_NUM_THREADS, MKL_NUM_THREADS according to trainer_count.
+       For OpenBLAS:
+         export OPENBLAS_NUM_THREADS, OPENBLAS_MAIN_FREE according to trainer_count. 
     '''
-    import platform
+    import platform, paddle
     if not platform.system() in ['Linux', 'Darwin']:
         return
 
@@ -103,16 +106,22 @@ def set_omp_mkl_env_vars(trainer_count):
 
     num_cores = num_physical_cores()
     num_processors = num_logical_processors()
-    if num_processors > num_cores:  # Hyper Threading is enabled
-        set_env("OMP_DYNAMIC", "true")
-        set_env("KMP_AFFINITY", "granularity=fine,compact,1,0")
-    else:
-        set_env("OMP_DYNAMIC", "false")
-        set_env("KMP_AFFINITY", "granularity=fine,compact,0,0")
+    if paddle.version.mkl() == 'ON':
+        if num_processors > num_cores:  # Hyper Threading is enabled
+            set_env("OMP_DYNAMIC", "true")
+            set_env("KMP_AFFINITY", "granularity=fine,compact,1,0")
+        else:
+            set_env("OMP_DYNAMIC", "false")
+            set_env("KMP_AFFINITY", "granularity=fine,compact,0,0")
     threads = num_processors / trainer_count
     threads = '1' if threads < 1 else str(threads)
-    set_env("OMP_NUM_THREADS", threads)
-    set_env("MKL_NUM_THREADS", threads)
+    if paddle.version.mkl() == 'ON':
+        set_env("OMP_NUM_THREADS", threads)
+        set_env("MKL_NUM_THREADS", threads)
+    else:
+        set_env("OPENBLAS_NUM_THREADS", threads)
+        if threads > 1:
+            set_env("OPENBLAS_MAIN_FREE", '1')
 
 
 def init(**kwargs):
@@ -129,7 +138,7 @@ def init(**kwargs):
     for key in args_dict.keys():
         args.append('--%s=%s' % (key, str(args_dict[key])))
 
-    set_omp_mkl_env_vars(kwargs.get('trainer_count', 1))
+    set_env_vars(kwargs.get('trainer_count', 1))
 
     if 'use_gpu' in kwargs:
         cp.g_command_config_args['use_gpu'] = kwargs['use_gpu']
diff --git a/python/setup.py.in b/python/setup.py.in
index 66ccfe808763d0e157f866ce08868e3fdebdea79..65ec58ecf98e693ecf02922129f6eef13cbe5303 100644
--- a/python/setup.py.in
+++ b/python/setup.py.in
@@ -31,6 +31,7 @@ patch           = '%(patch)d'
 rc              = '%(rc)d'
 istaged         = %(istaged)s
 commit          = '%(commit)s'
+with_mkl        = '%(with_mkl)s'
 
 def show():
     if istaged:
@@ -41,6 +42,9 @@ def show():
         print 'rc:', rc
     else:
         print 'commit:', commit
+
+def mkl():
+    return with_mkl
 '''
     commit = git_commit()
     with open(filename, 'w') as f:
@@ -51,7 +55,8 @@ def show():
             'rc': RC,
             'version': '${PADDLE_VERSION}',
             'commit': commit,
-            'istaged': ISTAGED})
+            'istaged': ISTAGED,
+            'with_mkl': '@WITH_MKL@'})
 
 write_version_py(filename='@PADDLE_SOURCE_DIR@/python/paddle/version.py')