1660ti系列以上显卡gpu版本pytorch安装指令

如果你想在拥有NVIDIA GeForce GTX 1660 Ti及以上显卡的系统上安装PyTorch，你需要首先确保你的CUDA和cuDNN库是最新的。以下是安装步骤：

1. **更新操作系统**:
- 对于Ubuntu/Debian: `sudo apt-get update && sudo apt-get upgrade`
- 对于 CentOS/Fedora: `sudo yum update`

2. **安装必要的库**:

   # 安装CUDA (替换`<version>`为你需要的CUDA版本)
   wget https://developer.nvidia.com/compute/cuda/<version>/repo/deb/x86_64/cuda-repo-deb_<version>_<os_name>-11-0-local.gpg.key
   sudo apt-key add cuda-repo-deb_<version>_<os_name>-11-0-local.gpg.key
   echo "deb [arch=compute_{arch}]{cuda_url}/stable-{os_name}/{cuda_version}-local /" | sudo tee /etc/apt/sources.list.d/cuda.list
   whereis libcudart.so > /dev/null || { echo "You need to install CUDA development package"; exit 1; }
   sudo apt-get update
   sudo apt-get install cuda

   # 安装cuDNN (同样替换版本号)
   sudo apt-get install libcudnn7 libcudnn7-dev

3. **安装PyTorch**:
- 首先，添加apt-pytorch-nightly源：

     curl https://raw.githubusercontent.com/pytorch/get-torch/main/install.sh | bash -s -- <CUDA_VERSION>

- 然后安装PyTorch：

     python3 -m torch.cuda.set_device 0  # 设定GPU设备为0
     python3 -m pip install torch torchvision torchaudio -f https://download.pytorch.org/whl/cpu/torch_stable.html  # CPU版本依赖
     python3 -m pip install torch torchvision torchaudio -f https://download.pytorch.org/whl/cu<cuda_version>/torch_stable.html  # GPU版本依赖

4. **检查安装**:

   python -c "import torch; print(torch.__version__, torch.cuda.is_available())"