A problem with GPU acceleration

Dear LAMMPS developers,
Now I am having a problem with GPU acceleration. I use the GPU package in LAMMPS. The server is equipped with Tesla K80, with 8 GPU cores. When I run three cases simultaneously as below, the system breaks down and restarts automatically. In detail, the system does not break down with only the first 2 cases running, and breaks down when I add the third case.

Dear LAMMPS developers,
    Now I am having a problem with GPU acceleration. I use the GPU package
in LAMMPS. The server is equipped with Tesla K80, with 8 GPU cores. When I
run three cases simultaneously as below, the system breaks down and
restarts automatically. In detail, the system does not break down with only
the first 2 cases running, and breaks down when I add the third case.

​if the machine is spontaneously restarting, it is highly likely a hardware
issue. e.g. you may not have sufficient power to drive those 8 GPUs all
under load.

axel.​

Thank you for your suggestion. The power is 1400W, and the engineer of K80 seller said that should be enough. Actually, when another student purely runs Lattice Boltzmann Method procedures on the same machine, with all 8 GPUs occupied and 99% GPU utilization according to command “nvidia-smi”, the machine never restarts spontaneously. That’s why I am wondering if this problem is related with MD algorithm.
With the above information, do you have any other suggestions?
Best wishes,
Huanhuan Tian