gpu - Cuda driver error 700 in call at file 'geryon/nvd_timer.h' in LAMMPS

Question

I am doing a relaxation calculation using gpu in a system of 800,000 atoms. Sometimes, I get the following error. Can you please tell me the cause and how to fix it?

Cuda driver error 700 in call at file 'geryon/nvd_timer.h' in line 76.
application called MPI_Abort(MPI_COMM_WORLD, -1) - process 0

Answer 1

简而言之，通过改变不同版本的CUDA和ARCH代码解决了。如果要使用2080，则必须选择-DGPU_ARCH=sm_70的cuda10.2。II 在 HPC 上有 3090 和 2080。我认为它是向下兼容的，但事实并非如此。我google了一下，有人说2080不支持cuda11.4。