How to Add and Run the example job on GPU compute node

Hi Team,
We have 1 Master and 1 Compute nodes. now i want to add GPU node(gpnode having 2 gpu cards i.e K40).
Please guide me

  1. how to add gup node and run the example job on it.
  2. can submit jobs on only gpu based/ cpu based jobs on GPU Node.
  3. does the jobs will submit based on gpu cards (card1, card2) - i am not clear with.

Please help me on this?

[root@master ~]# pbsnodes -aSj
                                                        mem       ncpus   nmics   ngpus
vnode           state           njobs   run   susp      f/t        f/t     f/t     f/t   jobs
--------------- --------------- ------ ----- ------ ------------ ------- ------- ------- -------
master          free                 0     0      0  126gb/126gb   16/16     0/0     0/0 --
compute-0-0     free                 0     0      0  126gb/126gb   16/16     0/0     0/0 --
[root@master ~]#


Adding a cpu node is same as adding a gpu node
qmgr -c “create note nodname”
If you want make PBS aware of those GPU cards, then you do with

  1. custom resources : Please check Basic GPU Scheduling AG-283
  2. cgroups hook

yes, while a gpu job would still need some cycles of cpu.

a. Yes if you use cgroups hook integration
b. or else you can customise using custom resources and CUDA_VISIBLE_DEVICE variable and assignment on the gpu node

Thank you