diff --git a/PaddleCV/yolov3/README.md b/PaddleCV/yolov3/README.md
index 58168bd6ee6acb25f0a180e11dcd74a26f921a42..32122acd7a440221f1c9d66616967b1bb8474520 100644
--- a/PaddleCV/yolov3/README.md
+++ b/PaddleCV/yolov3/README.md
@@ -163,7 +163,9 @@ Inference is used to get prediction score or image features based on trained mod
         --image_name=000000000139.jpg \
         --draw_threshold=0.5
 
-Inference speed:
+- Set ```export CUDA_VISIBLE_DEVICES=0``` to specifiy one GPU to infer.
+
+Inference speed（Tesla P40）:
 
 
 |   input size  | 608x608 | 416x416 | 320x320 |
diff --git a/PaddleCV/yolov3/README_cn.md b/PaddleCV/yolov3/README_cn.md
index e44bbbdb62a22b3105d26b7178804f6ecc3cccc0..b9a236bee877e28c183221e999adcade24cc944e 100644
--- a/PaddleCV/yolov3/README_cn.md
+++ b/PaddleCV/yolov3/README_cn.md
@@ -165,7 +165,9 @@ Train Loss
         --image_name=000000000139.jpg \
         --draw_threshold=0.5
 
-模型预测速度：
+- 通过设置export CUDA\_VISIBLE\_DEVICES=0指定单卡GPU预测。
+
+模型预测速度（Tesla P40）：
 
 
 |   input size  | 608x608 | 416x416 | 320x320 |