abacusmodeling / abacus-develop

An electronic structure package based on either plane wave basis or numerical atomic orbitals.
GNU Lesser General Public License v3.0
157 stars 165 forks source link

CUDA版本问题 #331

Closed SyntaxSmith closed 3 days ago

SyntaxSmith commented 2 months ago

Describe the bug

现在cuda版我观察到的有 1 好像多卡没用上,体系比较大的时候Mpirun -n 不能开大,所有进程占用显存会挤在第一张卡上 2 docker版似乎不能用OMP多线程,进程cpu占用率长期是100%,手动指定环境变量也无效,但是多进程会爆显存 3 conda版会报oom,xxmr2d:out of memory

Expected behavior

No response

To Reproduce

No response

Environment

No response

Additional Context

以上所有bug都是预编译版,我在请运维本地编译,我会在编译完成后重新测试。

Task list for Issue attackers (only for developers)

mohanchen commented 3 days ago

Thanks for your advice, would you like to propose a new issue on deepmodeling/abacus?(https://github.com/deepmodeling/abacus-develop) We will discuss issues there.