diff --git a/benchmark/paddle/image/run_openblas_infer.sh b/benchmark/paddle/image/run_openblas_infer.sh
index da034f3b9dff794e22086a5295ad2b0c2361c356..71a49231a5527ebee9f45d5f4650ce2a4f6a1c31 100755
--- a/benchmark/paddle/image/run_openblas_infer.sh
+++ b/benchmark/paddle/image/run_openblas_infer.sh
@@ -8,15 +8,19 @@ function clock_to_seconds() {
 }
 
 function infer() {
-  unset OMP_NUM_THREADS MKL_NUM_THREADS OMP_DYNAMIC KMP_AFFINITY
   topology=$1
   layer_num=$2
   bs=$3
-  thread=`nproc`
-  if [ $thread -gt $bs ]; then
-    thread=$bs
+  trainers=`nproc`
+  if [ $trainers -gt $bs ]; then
+    trainers=$bs
   fi
-  log="logs/infer-${topology}-${layer_num}-${thread}openblas-${bs}.log"
+  log="logs/infer-${topology}-${layer_num}-${trainers}openblas-${bs}.log"
+  threads=$((`nproc` / trainers))
+  if [ $threads -eq 0 ]; then
+    threads=1
+  fi
+  export OPENBLAS_NUM_THREADS=$threads
 
   models_in="models/${topology}-${layer_num}/pass-00000/"
   if [ ! -d $models_in ]; then
@@ -28,7 +32,7 @@ function infer() {
     --config="${topology}.py" \
     --use_mkldnn=False \
     --use_gpu=False \
-    --trainer_count=$thread \
+    --trainer_count=$trainers \
     --log_period=$log_period \
     --config_args="batch_size=${bs},layer_num=${layer_num},is_infer=True,num_samples=256" \
     --init_model_path=$models_in \
diff --git a/benchmark/paddle/image/run_openblas_train.sh b/benchmark/paddle/image/run_openblas_train.sh
index e9df83fee2a3f796b7234b39619364f6ee4d5dc9..935cff6f2c97d25d6de556cfee25e27dbe49b5b6 100755
--- a/benchmark/paddle/image/run_openblas_train.sh
+++ b/benchmark/paddle/image/run_openblas_train.sh
@@ -1,7 +1,7 @@
 set -e
 
 function train() {
-  unset OMP_NUM_THREADS MKL_NUM_THREADS OMP_DYNAMIC KMP_AFFINITY
+  export OPENBLAS_NUM_THREADS=1
   topology=$1
   layer_num=$2
   bs=$3
diff --git a/paddle/scripts/submit_local.sh.in b/paddle/scripts/submit_local.sh.in
index a94bc01b358c508132eb85920a2d4c0aa934dd51..8a352b0078d701f797f7202c85bd0e08201ac9b8 100755
--- a/paddle/scripts/submit_local.sh.in
+++ b/paddle/scripts/submit_local.sh.in
@@ -71,9 +71,7 @@ function threads_config() {
   # auto set OMP_NUM_THREADS and MKL_NUM_THREADS
   # according to trainer_count and total processors
   # only when MKL enabled
-  if [ "@WITH_MKL@" == "OFF" ]; then
-    return 0
-  fi
+  # auto set OPENBLAS_NUM_THREADS when do not use MKL
   processors=`grep "processor" /proc/cpuinfo|sort -u|wc -l`
   trainers=`grep -Eo 'trainer_count.[0-9]+' <<< "$@" |grep -Eo '[0-9]+'|xargs`
   if [ -z $trainers ]; then
@@ -83,12 +81,19 @@ function threads_config() {
   if [ $threads -eq 0 ]; then
     threads=1
   fi
-  if [ -z "$OMP_NUM_THREADS" ]; then
-    export OMP_NUM_THREADS=$threads
-  fi
-  if [ -z "$MKL_NUM_THREADS" ]; then
-    export MKL_NUM_THREADS=$threads
+  if [ "@WITH_MKL@" == "ON" ]; then
+    if [ -z "$OMP_NUM_THREADS" ]; then
+      export OMP_NUM_THREADS=$threads
+    fi
+    if [ -z "$MKL_NUM_THREADS" ]; then
+      export MKL_NUM_THREADS=$threads
+    fi
+  else
+    if [ -z "$OPENBLAS_NUM_THREADS" ]; then
+      export OPENBLAS_NUM_THREADS=$threads
+    fi
   fi
+  
 }
 
 PADDLE_CONF_HOME="$HOME/.config/paddle"
@@ -150,7 +155,7 @@ fi
 case "$1" in
     "train")
         threads_config $@
-        # echo $OMP_NUM_THREADS $MKL_NUM_THREADS
+        # echo $OMP_NUM_THREADS $MKL_NUM_THREADS $OPENBLAS_NUM_THREADS
         ${DEBUGGER} $PADDLE_BIN_PATH/paddle_trainer ${@:2}
         ;;
     "merge_model")