Windows11安装TensorFlow GPU失败：找不到nvidia-nccl-cu12==2.19.3版本

系统与CUDA版本 #

Windows 11系统，已安装CUDA 11.8，nvcc版本输出：

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2022 NVIDIA Corporation
Built on Wed_Sep_21_10:41:10_Pacific_Daylight_Time_2022
Cuda compilation tools, release 11.8, V11.8.89
Build cuda_11.8.r11.8/compiler.31833905_0

WSL2状态 #

已安装WSL2，当前状态：

NAME      STATE           VERSION

* Ubuntu    Stopped         2

执行pip install tensorflow[and-cuda]时出现以下错误：

Collecting tensorflow\[and-cuda\]
Using cached tensorflow-2.16.1-cp312-cp312-win_amd64.whl.metadata (3.5 kB)
Requirement already satisfied: tensorflow-intel==2.16.1 in c:\users\nguye\appdata\local\programs\python\python312\lib\site-packages (from tensorflow\[and-cuda\]) (2.16.1)
Collecting nvidia-cublas-cu12==12.3.4.1 (from tensorflow\[and-cuda\])
Using cached nvidia_cublas_cu12-12.3.4.1-py3-none-win_amd64.whl.metadata (1.5 kB)
Collecting nvidia-cuda-cupti-cu12==12.3.101 (from tensorflow\[and-cuda\])
Using cached nvidia_cuda_cupti_cu12-12.3.101-py3-none-win_amd64.whl.metadata (1.6 kB)
Collecting nvidia-cuda-nvcc-cu12==12.3.107 (from tensorflow\[and-cuda\])
Using cached nvidia_cuda_nvcc_cu12-12.3.107-py3-none-win_amd64.whl.metadata (1.5 kB)
Collecting nvidia-cuda-nvrtc-cu12==12.3.107 (from tensorflow\[and-cuda\])
Using cached nvidia_cuda_nvrtc_cu12-12.3.107-py3-none-win_amd64.whl.metadata (1.5 kB)
Collecting nvidia-cuda-runtime-cu12==12.3.101 (from tensorflow\[and-cuda\])
Using cached nvidia_cuda_runtime_cu12-12.3.101-py3-none-win_amd64.whl.metadata (1.5 kB)
Collecting nvidia-cudnn-cu12==8.9.7.29 (from tensorflow\[and-cuda\])
Using cached nvidia_cudnn_cu12-8.9.7.29-py3-none-win_amd64.whl.metadata (1.6 kB)
Collecting nvidia-cufft-cu12==11.0.12.1 (from tensorflow\[and-cuda\])
Using cached nvidia_cufft_cu12-11.0.12.1-py3-none-win_amd64.whl.metadata (1.5 kB)
Collecting nvidia-curand-cu12==10.3.4.107 (from tensorflow\[and-cuda\])
Using cached nvidia_curand_cu12-10.3.4.107-py3-none-win_amd64.whl.metadata (1.5 kB)
Collecting nvidia-cusolver-cu12==11.5.4.101 (from tensorflow\[and-cuda\])
Using cached nvidia_cusolver_cu12-11.5.4.101-py3-none-win_amd64.whl.metadata (1.6 kB)
Collecting nvidia-cusparse-cu12==12.2.0.103 (from tensorflow\[and-cuda\])
Using cached nvidia_cusparse_cu12-12.2.0.103-py3-none-win_amd64.whl.metadata (1.6 kB)
INFO: pip is looking at multiple versions of tensorflow\[and-cuda\] to determine which version is compatible with other requirements. This could take a while.
ERROR: Could not find a version that satisfies the requirement nvidia-nccl-cu12==2.19.3; extra == "and-cuda" (from tensorflow\[and-cuda\]) (from versions: 0.0.1.dev5)
ERROR: No matching distribution found for nvidia-nccl-cu12==2.19.3; extra == "and-cuda"

方案1：降级TensorFlow适配现有CUDA 11.8 #

• 卸载当前的CPU版TensorFlow：

  pip uninstall tensorflow-intel -y
  

• 安装适配CUDA 11.8的TensorFlow GPU版本（推荐2.14.0，对CUDA11.8支持完善）：

  pip install tensorflow==2.14.0
  

• 验证GPU可用性：运行以下Python代码

  import tensorflow as tf
  print(tf.config.list_physical_devices('GPU'))
  

  输出包含GPU设备信息即为安装成功。

方案2：升级CUDA到12.3并手动安装依赖 #

• 卸载系统中现有的CUDA 11.8，安装CUDA 12.3 Toolkit
• 先安装TensorFlow核心包：

  pip install tensorflow-intel==2.16.1
  

• 手动安装除NCCL外的所有CUDA依赖（Windows单GPU训练无需NCCL）：

  pip install nvidia-cublas-cu12==12.3.4.1 nvidia-cuda-cupti-cu12==12.3.101 nvidia-cuda-nvcc-cu12==12.3.107 nvidia-cuda-nvrtc-cu12==12.3.107 nvidia-cuda-runtime-cu12==12.3.101 nvidia-cudnn-cu12==8.9.7.29 nvidia-cufft-cu12==11.0.12.1 nvidia-curand-cu12==10.3.4.107 nvidia-cusolver-cu12==11.5.4.101 nvidia-cusparse-cu12==12.2.0.103
  

• 验证GPU可用性，步骤同方案1。

方案3：切换到WSL2 Ubuntu环境安装（推荐） #

• 启动WSL2 Ubuntu：

  wsl --start Ubuntu
  

• 在Ubuntu内安装Python 3.10或3.11（TensorFlow 2.16.1推荐版本，避开3.12兼容性问题）
• 直接执行安装命令：

  pip install tensorflow[and-cuda]
  

  Linux环境下nvidia-nccl-cu12对应版本存在，不会出现依赖缺失问题，且WSL2自动共享Windows的GPU驱动。

内容的提问来源于stack exchange，提问作者Phuc Nguyen