Skip to content

同样是cuda13,这个docker包只兼容v19.0、v19.1,不兼容v19.2、v19.3 #76558

@cz132023cz

Description

@cz132023cz

bug描述 Describe the Bug

v19.0的linux驱动包:NVIDIA-Linux-x86_64-580.65.06-grid.run
v19.1的linux驱动包:NVIDIA-Linux-x86_64-580.82.07-grid.run
v19.2的linux驱动包:NVIDIA-Linux-x86_64-580.95.05-grid.run
v19.3的linux驱动包:NVIDIA-Linux-x86_64-580.105.08-grid.run
主机底座的驱动是19.3的。
docker包:ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.2.2-gpu-cuda13.0-cudnn9.13

1、都是CUDA 13
系统是虚拟机飞牛1.0.0(debian、ubuntu可以同理)的esxi8.0u3g(pve同理),开vGPU,GPU是Nvidia Tesla A2
按照飞牛安装驱动的流程
https://www.cnblogs.com/amsilence/p/19086906

2、装完就装英伟达容器工具包(NVIDIA Container Toolkit)
3、然后下载 docker pull ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.2.2-gpu-cuda13.0-cudnn9.13
4、然后进去
docker run --gpus all --name paddle -it -v /vol1/1000/docker/paddle:/paddle ccr-2vdh3abv-pub.cnc.bj.baidubce.com/paddlepaddle/paddle:3.2.2-gpu-cuda13.0-cudnn9.13 /bin/bash

5、然后在docker里面检查
python -c "import paddle; print(paddle.version)"

结果一、v19.0和v19.1正常,只会提示3.2.2

结果二、v19.2和v19.3在提示3.2.2之前,还会提示
/usr/local/lib/python3.10/dist-packages/paddle/base/framework.py:824: UserWarning: You are using GPU version Paddle, but your CUDA device is not set properly. CPU device will be used by default.
warnings.warn(
3.2.2
也就是用不了GPU,只能降级为CPU

操作全部一样,就只有驱动版本的不同。

其他补充信息 Additional Supplementary Information

没有

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions