HPC Service

The UBDA HPC Platform provides researchers to run and execute their own programs and applications across multiple CPU cores, nodes and GPU environment with the parallel file system support. The HPC Platform uses job queue scheduling system for the user to run his/her programs/applications. Users require to submit jobs (PBS Script) into different job queues for different hardware resource, such as the requirement of the No. of computing nodes, No. of GPU Cards, and No. of CPU cores in a single computing node, etc.

To run a code on the UBDA HPC Platform (ubdaplatform.polyu.edu.hk), the following job queue can be selected as per your application requirements:

Job queue	Max no. of nodes	No. of CPU cores per node	No. of GPU card per node	Memory per node (GB)
q2s01	40	26	N/A	128
q4s01	3	64	N/A	1024
qgpu01	2	28	1	128
	4	32	2	128
	1	32	4	128
qmic01	2	68	N/A	128

Job queue	Max no. of CPU core for a job	Max no. concurrent job per user	Max no. of job(s) able to be submitted concurrently per user	Maximum run time limit (Walltime) for jobs (hrs)
q2s01	104	6	10	168
q4s01	104	6	10	168
qgpu01	104	6	10	168
qmic01	104	6	10	168

200 GB

For the different demands from different projects, users should let us know what computing resources they need, like No. of CPU cores, GPU, storage size as well as the applications to be setup on the UBDA HPC Platform. We are pleased to work with you to setup and build for running your intended programs/application. For such requirements, users should inform us of their plans so as to reserve the resources properly.

Singularity container software is installed to perform operating-system-level virtualization to run container image applications. Users are welcome to install any software into their home directory (/ubda/home/NetID) with general-user privilege, if there is something that you think should be made available to all users, email us and we will compile and install it for you. To see what is already installed, you can run "moduel avail" to get current available software list.

It is suggested to check the available of free nodes before job submission. Users can type the queue name directly in the SSH prompt to check the available free nodes, such as "q2s01" and "qgpu01" etc. Similarly, the node name ("ubda-d050", "ubda-d003" etc) is to report the CPU instant load and memory consumption. "checkquota" checks disk usage, "nvidia-smi-g003", "nvidia-smi-g012" etc check GPU card load. Other popular commands are "qub xxx.pbs", "qstat -na" and "qdel JobID" etc for job management.

UBDA HPC Platform uses PolyU NetID and Net Password to log in, users should keep their account private and should not share their UBDA account without prior approval.

For more information, please email to ubdae@polyu.edu.hk