diff --git a/deploy/pdserving/README.md b/deploy/pdserving/README.md
index b5d8953d73a97b0586fdc60de1c0531fb713f90c..34b525c47214d45819690ed7d161ce3d4dc0c322 100644
--- a/deploy/pdserving/README.md
+++ b/deploy/pdserving/README.md
@@ -156,7 +156,9 @@ The recognition model is the same.
 
     Multiple service requests can be sent at the same time if necessary.
 
-    The predicted performance data will be automatically written into the `PipelineServingLogs/pipeline.tracer` file:
+    The predicted performance data will be automatically written into the `PipelineServingLogs/pipeline.tracer` file.
+
+    Tested on 200 real pictures, and limited the detection long side to 960. The average QPS on T4 GPU can reach around 13:
 
     ```
     2021-05-12 10:03:24,812 ==================== TRACER ======================
diff --git a/deploy/pdserving/README_CN.md b/deploy/pdserving/README_CN.md
index 37d018d0ed2f95543fdf81ddb317de7542e5d244..4bb3427199ff60968226237bcf67277ae72a1b9d 100644
--- a/deploy/pdserving/README_CN.md
+++ b/deploy/pdserving/README_CN.md
@@ -153,7 +153,9 @@ python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_mobile_v2.0_rec_in
     ```
     有需要的话可以同时发送多个服务请求
 
-    预测性能数据会被自动写入 `PipelineServingLogs/pipeline.tracer` 文件中：
+    预测性能数据会被自动写入 `PipelineServingLogs/pipeline.tracer` 文件中。
+
+    在200张真实图片上测试，把检测长边限制为960。T4 GPU 上 QPS 均值可达到13左右：
 
     ```
     2021-05-12 10:03:24,812 ==================== TRACER ======================
@@ -191,6 +193,8 @@ python3 -m paddle_serving_client.convert --dirname ./ch_ppocr_mobile_v2.0_rec_in
     2021-05-12 10:03:24,910         chl2(In: ['rec'], Out: ['@DAGExecutor']) size[0/0]
     ```
 
+
+
 <a name="FAQ"></a>
 ## FAQ
 **Q1**： 发送请求后没有结果返回或者提示输出解码报错