<h3>检查cuda版本</h3><div><pre><code>nvcc -V</code></pre></div><p><img src="https://img.ithere.net/article/article/image/2024/6/14/CqLgHnafkAduwcXENObFRNEHfAFjKsjx.png" alt="image.png"><img alt=""><br></p><h3>检查nvidia驱动</h3><div><pre><code>nvidia-smi</code></pre></div><p><img src="https://img.ithere.net/article/article/image/2024/6/14/LpjXLoftXwOjOzJJtWjvJLoomTOrSYDZ.png" alt="image.png"><img alt=""><br></p><p><br></p><p>cuda版本和驱动是兼容的，但是执行torch.cuda.is_available()报错NumCudaDevices()</p><p>需要安装nvidia-fabricmanager-535</p><ul><li><p>nvidia-fabricmanager版本需要和驱动版本完全一致，驱动版本是535.104.05，nvidia-fabricmanager版本也得是这个</p><p>去<a href="https://developer.download.nvidia.cn/compute/cuda/repos/ubuntu2204/x86_64/">这里</a>下载与驱动版本完全一致的nvidia-fabricmanager-535_535.104.05-1_amd64.deb</p></li></ul><div><pre><code>#若有旧的版本，请删去后重新下载
#手动安装
sudo apt-get install ./nvidia-fabricmanager-535_535.104.05-1_amd64.deb
#解除禁用
sudo systemctl enable nvidia-fabricmanager
#重启
sudo systemctl restart nvidia-fabricmanager
#检查状态
sudo systemctl status nvidia-fabricmanager
#安装成功

python3 -c "import torch; print(torch.cuda.is_available())"</code></pre></div><p><img src="https://img.ithere.net/article/article/image/2024/6/14/tkSTuEeuqoXyrlvEEDFOkLJXtzEcvYmf.png" alt="image.png"><img alt=""><br></p><p>显示可用</p>

解决torch.cuda.is_available()报错NumCudaDevices()

检查cuda版本
nvcc -V


检查nvidia驱动
nvidia-smi



cuda版本和驱动是兼容的，但是执行torch.cuda.is_available()报错NumCudaDevices()
需要安装nvidia-fabricmanager-535


nvidia-fabricmanager版本需要和驱动版本完全一致，驱动版本是535.104.05，nvidia-fabricmanager版本也得是这个
去这里下载与驱动版本完全一致的nvidia-fabricmanager-535_535.104.05-1_amd64.deb


#若有旧的版本，请删去后重新下载
#手动安装
sudo apt-get install ./nvidia-fabricmanager-535_535.104.05-1_amd64.deb
#解除禁用
sudo systemctl enable nvidia-fabricmanager
#重启
sudo systemctl restart nvidia-fabricmanager
#检查状态
sudo systemctl status nvidia-fabricmanager
#安装成功

python3 -c "import torch; print(torch.cuda.is_available())"


显示可用