lammps编译GPU加速失败求助

adong

使用cmake编译GPU加速的lammps，用nohup mpirun -np 12 lmp_gpu -sf gpu -pk gpu 1 -in system.in &提交任务，使用[size=0.85em]nvidia-smi发现GPU有12个同样的任务在运行，GPU占用很小，请问有人遇到这种问题吗？该如何解决？
而如果使用nohup mpirun -np 1 lmp_gpu -sf gpu -pk gpu 1 -in system.in &提交任务就直接报错：Cuda driver error 1 in call at file '/home/hnu/software/lammps-stable_29Sep2021_update1/lib/gpu/geryon/nvd_kernel.h' in line 379.
application called MPI_Abort(MPI_COMM_WORLD, -1) - process 0
-----GPU 驱动正常----
$nvcc -V
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2019 NVIDIA Corporation
Built on Fri_Feb__8_19:08:17_PST_2019
Cuda compilation tools, release 10.1, V10.1.105

$ nvidia-smi
Sat Dec  4 21:14:42 2021      
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 460.32.03    Driver Version: 460.32.03    CUDA Version: 11.2     |
|-------------------------------+----------------------+----------------------+

abin

先把Display Driver 与 cuda 版本搞一致再测试.

adong

abin 发表于 2021-12-4 21:20
先把Display Driver 与 cuda 版本搞一致再测试.

好的，谢谢老师