About NU Research Computing¶
Nazarbayev University Research Computing (NU RC) operates shared high-performance computing systems for research and teaching. The clusters provide more CPU cores, memory, accelerators and storage than a workstation, but they are not a faster replacement for every desktop task.
Choose HPC when you need¶
- a parallel application that can use many CPU cores or GPUs;
- more memory than a workstation provides;
- many independent runs, samples or parameter combinations;
- long, unattended batch execution;
- a reproducible software environment shared by a research group.
For latency-sensitive services, continuous data ingestion, web applications or guaranteed real-time response, use a dedicated server or an appropriate cloud/service platform. Slurm jobs wait in a queue and therefore have no guaranteed start time.
How the service works¶
- Connect to a login node to transfer files, edit scripts, compile code and submit work.
- Describe the required resources in a Slurm batch script.
- Slurm places the job in a queue and starts it on one or more compute nodes when resources are available.
- The application writes results and logs to files; use accounting data to tune the next request.
Login nodes are shared
Do not run production computations, large data analyses or GPU workloads on a login node. Use a batch job or a short Slurm interactive allocation.
Start here¶
- Submit a first two-minute job
- Choose a cluster and partition
- Adapt a serial, OpenMP, MPI or GPU template
- Find software and create environments
- Check quotas, time limits and responsibilities
Expectations for reproducible work¶
Record the following with each project:
- the Git commit of your source and batch scripts;
- exact module names from
module list; - environment files such as
environment.ymlorrequirements.txt; - the Slurm job ID and relevant
sacctoutput; - input-data versions and checksums where practical.
This information makes a calculation easier to reproduce and gives the support team the evidence needed to diagnose failures.
Support¶
For accounts, access, software installation and quota requests, use the NU Helpdesk service referenced by the university. For technical questions, contact hpcadmin@nu.edu.kz and include the cluster, job ID, batch script, module list and the smallest relevant error excerpt. Never email passwords, private keys or access tokens.