diff --git a/doc/RUN_IN_DOCKER.md b/doc/RUN_IN_DOCKER.md
index edfd914123b2262b36e85552b179a0e412b68413..c111f884e376556193b2f5574fdd818f50e13b6d 100644
--- a/doc/RUN_IN_DOCKER.md
+++ b/doc/RUN_IN_DOCKER.md
@@ -135,6 +135,18 @@ pip install paddle-serving-server-gpu
 
 ### Test example
 
+When running the GPU Server, you need to set the GPUs used by the prediction service. By default, the GPU with index 0 is used. You can configure it in two ways:
+
+1. Using the `CUDA_VISIBLE_DEVICES` environment variable, the following example specifies two GPUs with an index of 0 and 1:
+
+   ```shell
+    export CUDA_VISIBLE_DEVICES=0,1
+   ```
+
+2. Using the `--gpu_ids` option, which will overrides the configuration of `CUDA_VISIBLE_DEVICES`.
+
+
+
 Get the trained Boston house price prediction model by the following command:
 
 ```bash
diff --git a/doc/RUN_IN_DOCKER_CN.md b/doc/RUN_IN_DOCKER_CN.md
index 47914a2fdd6aa78956b783ab813edb5d3a587b7d..cf70a8be327b119c039250c7bd95155df1160a57 100644
--- a/doc/RUN_IN_DOCKER_CN.md
+++ b/doc/RUN_IN_DOCKER_CN.md
@@ -127,11 +127,17 @@ pip install paddle-serving-server-gpu
 
 ### 测试example
 
-GPU版本在运行Server端代码前需要设置`CUDA_VISIBLE_DEVICES`环境变量来指定预测服务使用的GPU，下面的示例为指定索引为0和1两块GPU：
+在运行GPU版Server时需要设置预测服务使用的GPU，缺省状态默认使用索引为0的GPU。可以通过下面两种方式进行配置：
+
+1. 使用`CUDA_VISIBLE_DEVICES`环境变量，下面的示例为指定索引为0和1两块GPU：
+
+   ```shell
+    export CUDA_VISIBLE_DEVICES=0,1
+   ```
+
+2. 使用`--gpu_ids`选项，该选项将覆盖`CUDA_VISIBLE_DEVICES`的配置。
+
 
-```bash
- export CUDA_VISIBLE_DEVICES=0,1
-```
 
 通过下面命令获取训练好的Boston房价预估模型：
 
@@ -145,7 +151,7 @@ tar -xzf uci_housing.tar.gz
   在Server端（容器内）运行：
 
   ```bash
-  python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --name uci 
+  python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --name uci --gpu_ids 0
   ```
 
   在Client端（容器内或容器外）运行：
@@ -159,7 +165,7 @@ tar -xzf uci_housing.tar.gz
   在Server端（容器内）运行：
 
   ```bash
-  python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292
+  python -m paddle_serving_server_gpu.serve --model uci_housing_model --thread 10 --port 9292 --gpu_ids 0
   ```
 
   在Client端（容器内或容器外，需要安装`paddle-serving-client`包）运行下面Python代码：