In order for a job to be assigned to a GPU node, the partition gpu must be chosen:

#SBATCH --partition=gpu

In addition, the GPU cards which are needed by a job must be requested in the job script as so-called generic resources:

#SBATCH --gres=gpu:n

where n is the number of cards needed. So if 2 cards are need this would become:

#SBATCH --gres=gpu:2

Note that some nodes have 2 GPU cards whereas others have 4. Therefore you should in general not request more than 1/4 of the total number of cores, i.e. not more than 8 cores per card. Correspondingly you should normally not request more than 1/4 of the main memory per GPU card.