nvidia errors in TensorFlow

最新推荐文章于 2024-05-29 08:09:47 发布

yusisc

最新推荐文章于 2024-05-29 08:09:47 发布

阅读量177

点赞数

分类专栏： python

本文链接：https://blog.youkuaiyun.com/yusisc/article/details/81364795

版权

python 专栏收录该内容

22 篇文章

订阅专栏

本文探讨了在使用TensorFlow-GPU时遇到NVIDIA错误的情况，并提供了如何检查GPU内存占用的方法，通过nvidia-smi命令来定位问题所在。特别提到PyCharm科学模式下进程不会自动退出导致的问题。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

Go to check GPU memory usage, when encountering nvidia or cuda error.

I’ve been worked with tensorflow-gpu for a while. When run the codes, nvidia reports kinds of errors without clear instruction some time.

As far as my knowledge, met a nvidia error, one should check whether the gpu memory is occupied by other process by

nvidia-smi -l

As the following screenshot shows, the GPU-util is 0%, but the memory is nearly used out. If one run a new tensorflow process, errors are likely to be reported.
If you are coding with pycharm, there is scitific mode. In this mode, the process will not exit automatically, so the tensorflow session remains consuming the GPU memrory.

这里写图片描述