diff --git a/fluid/neural_machine_translation/transformer/.run_ce.sh b/fluid/neural_machine_translation/transformer/.run_ce.sh
index e37856e5055ac3689416355ca9b29e6b8911598f..50161ca4662547dd19a098ee1f360a717da1b9a7 100644
--- a/fluid/neural_machine_translation/transformer/.run_ce.sh
+++ b/fluid/neural_machine_translation/transformer/.run_ce.sh
@@ -24,4 +24,12 @@ train(){
         dropout_seed 10
 }
 
+cudaid=${transformer:=0} # use 0-th card as default
+export CUDA_VISIBLE_DEVICES=$cudaid
+
+train | python _ce.py
+
+cudaid=${transformer_m:=0,1,2,3} # use 0,1,2,3 card as default
+export CUDA_VISIBLE_DEVICES=$cudaid
+
 train | python _ce.py
diff --git a/fluid/neural_machine_translation/transformer/_ce.py b/fluid/neural_machine_translation/transformer/_ce.py
index b774af075b028f4afafeb575f65cc8c9de8dc96b..49e67a63c8377d01ff196ccb49a2e3f93814bcc0 100644
--- a/fluid/neural_machine_translation/transformer/_ce.py
+++ b/fluid/neural_machine_translation/transformer/_ce.py
@@ -7,14 +7,22 @@ from kpi import CostKpi, DurationKpi, AccKpi
 
 #### NOTE kpi.py should shared in models in some way!!!!
 
-train_cost_kpi = CostKpi('train_cost', 0.02, 0, actived=True)
-test_cost_kpi = CostKpi('test_cost', 0.005, 0, actived=True)
-train_duration_kpi = DurationKpi('train_duration', 0.06, 0, actived=True)
+train_cost_card1_kpi = CostKpi('train_cost_card1', 0.01, 0, actived=True)
+test_cost_card1_kpi = CostKpi('test_cost_card1', 0.005, 0, actived=True)
+train_duration_card1_kpi = DurationKpi(
+    'train_duration_card1', 0.06, 0, actived=True)
+train_cost_card4_kpi = CostKpi('train_cost_card4', 0.01, 0, actived=True)
+test_cost_card4_kpi = CostKpi('test_cost_card4', 0.005, 0, actived=True)
+train_duration_card4_kpi = DurationKpi(
+    'train_duration_card4', 0.06, 0, actived=True)
 
 tracking_kpis = [
-    train_cost_kpi,
-    test_cost_kpi,
-    train_duration_kpi,
+    train_cost_card1_kpi,
+    test_cost_card1_kpi,
+    train_duration_card1_kpi,
+    train_cost_card4_kpi,
+    test_cost_card4_kpi,
+    train_duration_card4_kpi,
 ]
 
 
diff --git a/fluid/neural_machine_translation/transformer/train.py b/fluid/neural_machine_translation/transformer/train.py
index beb2b7d3e6e6db09ac9269cd285f71446bb69270..1148fe58845003f44eb80186ecd3b462c130c209 100644
--- a/fluid/neural_machine_translation/transformer/train.py
+++ b/fluid/neural_machine_translation/transformer/train.py
@@ -463,9 +463,9 @@ def train_loop(exe, train_progm, dev_count, sum_cost, avg_cost, lr_scheduler,
                          "pass_" + str(pass_id) + ".infer.model"),
             data_input_names[:-2] + util_input_names, [predict], exe)
     if args.enable_ce:  # For CE
-        print("kpis\ttrain_cost\t%f" % total_avg_cost)
-        print("kpis\ttest_cost\t%f" % val_avg_cost)
-        print("kpis\ttrain_duration\t%f" % time_consumed)
+        print("kpis\ttrain_cost_card%d\t%f" % (dev_count, total_avg_cost))
+        print("kpis\ttest_cost_card%d\t%f" % (dev_count, val_avg_cost))
+        print("kpis\ttrain_duration_card%d\t%f" % (dev_count, time_consumed))
 
 
 def train(args):