I send with:
srun -n 1 -N 1 --gres=gpu:2 mpirun executable and it´s works, but only I can 
send with nodo compute0-0, with compute-0-1 and compute-0-2 I get the error. 

srun: error: Unable to allocate resources: Requested node configuration is not 
available

I don’t know what I missed because I have the same configuration in all nodes.

Any idea? Thanks

-----Mensaje original-----
De: Peter Kjellström [mailto:[email protected]] 
Enviado el: jueves, 01 de octubre de 2015 4:03
Para: slurm-dev
CC: slurm-dev
Asunto: [slurm-dev] Re: RV: I can´t send job with gpu


On Wed, 30 Sep 2015 12:47:28 -0700
Fany Pagés Díaz <[email protected]> wrote:

> I have a test cluster with three nodes with two gpus each one. In each 
> node this installed cudatoolkit-3.2.16.It´s works fine with mpirun but 
> with slurm I have the next error. what I missed?
> 
>  
> 
> [root@cluster bin]# srun -n5 -N 3 --gres=gpu:2 ejemplocudampi
> 
> srun: error: Unable to allocate resources: Requested node 
> configuration is not available

Maybe a good first step is to figure out which part of your spec doesn't exist. 
Try each on its own:

 -n1 --gres=gpu:2

and

 -n5 -N3

Also, your compute nodes seem to be setup with 8 sockets and 1 core per 
socket...

/Peter K


Reply via email to