Ubuntu22.04安装CUDA Toolkit12.6后nvcc版本仍为11.5的解决方法

我在Ubuntu 22.04系统中，按照官方指引执行以下命令安装了CUDA Toolkit 12.6 Update 1：

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/x86_64/cuda-ubuntu2204.pin
sudo mv cuda-ubuntu2204.pin /etc/apt/preferences.d/cuda-repository-pin-600
wget https://developer.download.nvidia.com/compute/cuda/12.6.1/local_installers/cuda-repo-ubuntu2204-12-6-local_12.6.1-560.35.03-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu2204-12-6-local_12.6.1-560.35.03-1_amd64.deb
sudo cp /var/cuda-repo-ubuntu2204-12-6-local/cuda-*-keyring.gpg /usr/share/keyrings/
sudo apt-get update
sudo apt-get -y install cuda-toolkit-12-6

安装成功后，nvcc版本仍为11.5。系统中同时安装了两个相关包：

$ dpkg -l nvidia-cuda-toolkit
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name                Version         Architecture Description
+++-===================-===============-============-=================================
ii  nvidia-cuda-toolkit 11.5.1-1ubuntu1 amd64        NVIDIA CUDA development toolkit

$ dpkg -l cuda-toolkit-12-6
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name              Version      Architecture Description
+++-=================-============-============-=================================
ii  cuda-toolkit-12-6 12.6.1-1     amd64        CUDA Toolkit 12.6 meta-package

nvcc -V输出：

$ nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2021 NVIDIA Corporation
Built on Thu_Nov_18_09:45:30_PST_2021
Cuda compilation tools, release 11.5, V11.5.119
Build cuda_11.5.r11.5/compiler.30672275_0

查看cuda-toolkit-12-6内容，发现不含nvcc：

$ dpkg -L cuda-toolkit-12-6
/.
/usr
/usr/local
/usr/local/cuda-12.6
/usr/local/cuda-12.6/version.json
/usr/share
/usr/share/doc
/usr/share/doc/cuda-toolkit-12-6
/usr/share/doc/cuda-toolkit-12-6/changelog.Debian.gz

nvidia-smi输出：

$ nvidia-smi
Mon Sep 23 13:11:55 2024       
+---------------------------------------------------------------------------------------+
| NVIDIA-SMI 535.183.01             Driver Version: 535.183.01   CUDA Version: 12.2     |
|-----------------------------------------+----------------------+----------------------+
| GPU  Name                 Persistence-M | Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp   Perf          Pwr:Usage/Cap |         Memory-Usage | GPU-Util  Compute M. |
|                                         |                      |               MIG M. |
|=========================================+======================+======================|
|   0  NVIDIA GeForce RTX 4090        Off | 00000000:81:00.0 Off |                  Off |
|  0%   24C    P8              21W / 450W |  10893MiB / 24564MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
|   1  NVIDIA GeForce RTX 4090        Off | 00000000:C1:00.0 Off |                  Off |
|  0%   24C    P8              32W / 450W |    818MiB / 24564MiB |      0%      Default |
|                                         |                      |                  N/A |
+-----------------------------------------+----------------------+----------------------+
                                                                                         
+---------------------------------------------------------------------------------------+
| Processes:                                                                            |
|  GPU   GI   CI        PID   Type   Process name                            GPU Memory |
|        ID   ID                                                             Usage      |
|=======================================================================================|
|    0   N/A  N/A      1720      G   /usr/lib/xorg/Xorg                            4MiB |
|    0   N/A  N/A      2174      C   /usr/local/bin/ollama                       384MiB |
|    0   N/A  N/A   1808406      C   /opt/tljh/user/bin/python                 10486MiB |
|    1   N/A  N/A      1720      G   /usr/lib/xorg/Xorg                           15MiB |
|    1   N/A  N/A      1896      G   /usr/bin/gnome-shell                         10MiB |
|    1   N/A  N/A      2174      C   /usr/local/bin/ollama                       384MiB |
|    1   N/A  N/A   1808406      C   /opt/tljh/user/bin/python                   386MiB |
+---------------------------------------------------------------------------------------+

请问如何将nvcc版本更新至12.6？

1. 安装CUDA 12.6的编译器组件 #

cuda-toolkit-12-6只是元包，不包含nvcc等核心编译工具。需要单独安装对应版本的编译器包：

sudo apt-get install cuda-nvcc-12-6

安装完成后，12.6版本的nvcc会被放在/usr/local/cuda-12.6/bin/nvcc路径下。

2. 调整环境变量优先级 #

当前系统优先调用旧版本的nvcc，需要让CUDA 12.6的路径排在系统路径前面：

• 临时生效（仅当前终端）：
  执行以下命令：

export PATH=/usr/local/cuda-12.6/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-12.6/lib64:$LD_LIBRARY_PATH

随后验证版本：

nvcc -V

此时应该显示12.6版本信息。

• 永久生效（所有终端会话）：
  编辑用户shell配置文件（比如~/.bashrc，若使用zsh则编辑~/.zshrc）：

nano ~/.bashrc

在文件末尾添加：

export PATH=/usr/local/cuda-12.6/bin:$PATH
export LD_LIBRARY_PATH=/usr/local/cuda-12.6/lib64:$LD_LIBRARY_PATH

保存退出后，重新加载配置：

source ~/.bashrc

3. 可选：卸载旧版本CUDA Toolkit #

如果不需要保留11.5版本，可以彻底卸载避免路径冲突：

sudo apt-get remove --purge nvidia-cuda-toolkit
sudo apt-get autoremove
sudo apt-get clean

4. 验证驱动兼容性 #

当前显卡驱动版本535.183.01最高支持CUDA 12.2，而你安装的是CUDA 12.6。虽然CUDA向前兼容，但如果出现兼容性问题，建议升级驱动到对应版本：

sudo apt-get install nvidia-driver-560

安装完成后重启系统，再通过nvidia-smi和nvcc -V验证版本匹配情况。

内容的提问来源于stack exchange，提问作者Raptor