diff --git a/paddle/fluid/framework/downpour_worker.cc b/paddle/fluid/framework/downpour_worker.cc
index d590986841d817538e9d1f6e73dd8777172312a1..28ea5f55acbf53ad4dcb95422197c6f6ad131979 100644
--- a/paddle/fluid/framework/downpour_worker.cc
+++ b/paddle/fluid/framework/downpour_worker.cc
@@ -188,6 +188,7 @@ void DownpourWorker::TrainFilesWithProfiler() {
   double push_dense_time = 0.0;
   int cur_batch;
   int batch_cnt = 0;
+  uint64_t total_inst = 0;
   timeline.Start();
   while ((cur_batch = device_reader_->Next()) > 0) {
     timeline.Pause();
@@ -315,6 +316,7 @@ void DownpourWorker::TrainFilesWithProfiler() {
     }
 
     thread_scope_->DropKids();
+    total_inst += cur_batch;
     ++batch_cnt;
 
     if (thread_id_ == 0) {
@@ -326,6 +328,7 @@ void DownpourWorker::TrainFilesWithProfiler() {
         }
         fprintf(stderr, "mean read time: %fs\n", read_time / batch_cnt);
         fprintf(stderr, "IO percent: %f\n", read_time / total_time * 100);
+        fprintf(stderr, "%6.2f instances/s\n", total_inst / total_time);
       }
     }
     timeline.Start();
diff --git a/paddle/fluid/framework/hogwild_worker.cc b/paddle/fluid/framework/hogwild_worker.cc
index 4c51067abfba6b077d6f3484648e21c58cd43b13..e902a5781e3578db4bbe70f01c692e030dfcdfab 100644
--- a/paddle/fluid/framework/hogwild_worker.cc
+++ b/paddle/fluid/framework/hogwild_worker.cc
@@ -89,6 +89,7 @@ void HogwildWorker::TrainFilesWithProfiler() {
   int cur_batch;
   int batch_cnt = 0;
   timeline.Start();
+  uint64_t total_inst = 0;
   while ((cur_batch = device_reader_->Next()) > 0) {
     VLOG(3) << "read a batch in thread " << thread_id_;
     timeline.Pause();
@@ -101,6 +102,7 @@ void HogwildWorker::TrainFilesWithProfiler() {
       op_total_time[i] += timeline.ElapsedSec();
       total_time += timeline.ElapsedSec();
     }
+    total_inst += cur_batch;
     ++batch_cnt;
     thread_scope_->DropKids();
     if (thread_id_ == 0) {
@@ -111,6 +113,7 @@ void HogwildWorker::TrainFilesWithProfiler() {
         }
         fprintf(stderr, "mean read time: %fs\n", read_time / batch_cnt);
         fprintf(stderr, "IO percent: %f\n", read_time / total_time * 100);
+        fprintf(stderr, "%6.2f instances/s\n", total_inst / total_time);
       }
     }
     timeline.Start();