1. Download the source
Official site, but the download link is broken down, when I visited it. (We can go directly to pypi site to download the source)
# tar the source
tar xfz pycuda-VERSION.tar.gz
2. Build
2.1 NVCC
When I build the project, I faced error with “NVCC”. we could ignore it, and we can build successfully, while it will greatly limit the functions.
So we should first source correctly the NVCC library to PATH.
export PATH=$PATH:/usr/local/cuda-10.2/bin
To find your cuda path (for me it is “/usr/local/cuda-10.2”) :
find / -type d -name cuda 2>/dev/null
Then add NVCC path to the complier.py file
de

本文档详细介绍了如何在Linux上安装PyCuda,包括从官方源下载(链接失效)到编译步骤,特别是解决NVCC错误的方法,以及如何测试安装是否成功。
最低0.47元/天 解锁文章
1428





