新环境下安装pytorch gpu

电脑显卡状态

nvidia-smi

Pasted image 20250404103958

CUDA下载

官网链接 链接: https://developer.nvidia.com/cuda-toolkit-archive
image.png#pic_center
验证

nvcc -V

image.png#pic_center

在官网下载pytorch

官网链接链接: Start Locally | PyTorch
image.png#pic_center

测试pytorch gpu是否可用

import torch  
print(torch.__version__)  
print(torch.version.cuda)  
print(torch.cuda.is_available())  #输出为True，则安装无误  
print(torch.cuda.current_device())  
print(torch.cuda.get_device_name(0))  
print(torch.cuda.get_device_capability(0))  
import os  
import subprocess  
cuda_home = os.environ.get('CUDA_HOME')  
if cuda_home:  
    nvcc_path = os.path.join(cuda_home, 'bin', 'nvcc.exe')  
    if os.path.exists(nvcc_path):  
        output = subprocess.check_output([nvcc_path, '-V'], universal_newlines=True)  
        print(output)  
    else:  
        print("nvcc not found in CUDA_HOME/bin")  
else:  
    print("CUDA_HOME is not set")

image.png#pic_center