Can't find CUDA when I use CUDA10

[gongweibao]

when I use CUDA10, it reports that serving can't find CUDA and I use CUDA 9 resolve it!

[github-actions]

Message that will be displayed on users' first issue

[barrierye]

CUDA9 is used by default for Paddle Serving.
If you need CUDA10, you can compile it from source.

[gongweibao]

Why not just release CUDA-9.0, CUDA-10.0 version's wheel package?

[barrierye]

We will consider release CUDA9 and CUDA10 version later.

[guru4elephant]

will release in 0.4.0

[sevenold]

@guru4elephant

example: https://github.com/PaddlePaddle/Serving/tree/develop/python/examples/ocr

paddle-serving-app        0.1.1                    pypi_0    pypi
paddle-serving-client     0.3.1                    pypi_0    pypi
paddle-serving-server     0.3.1                    pypi_0    pypi
paddle-serving-server-gpu 0.3.1                    pypi_0    pypi
paddlepaddle-gpu          1.8.3.post97             pypi_0    pypi

[barrierye]

Please run ldd <PATH/TO/SERVING_BIN>| grep cudart to see the output.

By the way, CUDA10 has been supported recently. You can download CUDA10 temporary version binary file(https://paddle-serving.bj.bcebos.com/temporary_files_for_docker/serving-gpu-0.3.1.tar.gz) and use it with the
command export SERVING_BIN=<PATH/TO/CUDA10_SERVING_BIN>.

[sevenold]

• docker image : paddlepaddle/serving:latest-cuda10.0-cudnn7
• cudart

[root@PC models]# ldd $SERVING_BIN | grep cudart
        libcudart.so.10.0 => /usr/local/cuda-10.0/targets/x86_64-linux/lib/libcudart.so.10.0 (0x00007f215c489000)

@barrierye

[barrierye]

Please check nvidia devices by ls /dev/ | grep nvidia

[iceriver97]

您好，在AIstudio上运行Serving的相关项目也需要更改CUDA的版本吗？