3. Using GPU

1. Login to the System with GPU.

KAGAYAKI system is available for GPU. (2022.3)

ssh kagayaki -l username

1. Interactive Use

Specify the required number of GPUs and submit your jobs with the GPU queue.* For the details of the queue, please refer the page of computing server *

1-1. Submitting Job
GPU x1:

qsub -q GPU-1 -I -lselect=1:ngpus=1

GPU x2:

qsub -q GPU-S -I -lselect=2:ngpus=1

GPU x4(Program must support MPI or other inter-node communication):

qsub -q GPU-L -I -lselect=4:ngpus=1

1-2. Confirmation

Are you seeing the specified number of GPUs?


1-3. CUDA

CUDA module is required for using CUDA library.
Notice: CUDA module is available only after qsub command shown above.

module load cuda/11.6

2. Batch Job

Run the job via Job Script.
Firstly, copy the sample file to your home directory.

cp /Samples/GPU/* .

2-1. Let's try to run the job with job script.


2-2. Confirm the job result made on the current directory.

cat nvidia-smi.log