Skip to content

Cluster Maintenance Updates and News

  • 1 min read

MIG GPU Testing on Wulver Now Available

We’re excited to announce that MIG-enabled GPUs are now available on Wulver for workflow testing!

We currently have 4 GPUs configured with MIG (Multi-Instance GPU) profiles as follows:

  • 40 GB profile – 1 MIG instance per GPU
  • 20 GB profile – 1 MIG instance per GPU
  • 10 GB profile – 2 MIG instances per GPU

This effectively allows the 4 GPUs to perform as 16 GPUs of varying RAM sizes. These MIG instances allow you to run multiple workloads in parallel with dedicated GPU resources, improving efficiency for smaller jobs and testing scenarios.

Who can use them?

All Wulver users are welcome to test their workflows on these new MIG profiles. Using the debug_gpu partition no Service Units (SUs) will be charged for these jobs. Modify your batch scripts to include these directives:

#SBATCH --partition=debug_gpu
#SBATCH --qos=debug
#SBATCH --gres=gpu:a100_10g:1      # Change to 20g or 40g as needed
#SBATCH --time=59:00 # Debug_gpu partition has a 12 hour walltime limit
What should you do?
  • Read through the MIG documentation on the HPC website.
  • Test your GPU-enabled workflows with these MIG resources.
  • Verify that your job scripts and containers handle MIG devices correctly.
  • Share feedback on performance and any issues you encounter.

This is a testing phase, so configurations may change based on usage and feedback. For more details, check the MIG documentation.

  • 1 min read

Wulver Outage

As part of NJIT's migration to a new Virtual Machine (VM) platform, Nutanix from (now ultra expensive) VMWare, Wulver will undergo an unplanned but required downtime to migrate critical virtual infrastructure hosting head, login, Open OnDemand and Slurm nodes starting at 8:00 AM on Friday, August 29. The cluster will be unavailable until all migration work is completed.

  • Expected duration: Up to 12 hours (work may finish sooner)
  • Reason: Migration to the Nutanix VM platform

Important Information:

  • Any jobs submitted before the outage that would not finish in time will be held in the queue and will resume after the cluster is back online. Please plan your usage and submissions accordingly.
  • There is a minor risk that queued jobs will be lost during migration. We will monitor this and inform affected users, if necessary.
  • Updates will be provided if there is any change to the expected downtime window.

We apologize for any inconvenience and appreciate your understanding as we make this important upgrade.

  • 1 min read

Wulver Maintenance

Wulver will be out of service on Tuesday, September 9th, for OS and SLURM updates.

Maintenance Plans

  • Upgrade the OS: Upgrade from RHEL 8 to RHEL 9: This will resolve the glibc error users encounter when compiling the latest packages. For details, see FAQ.
  • Upgrade SLURM version.
  • Implement new SU calculation: This will allow users to consume fewer SUs when using a single GPU instead of all 4 GPUs on a node.
  • Implement MIG.
  • Add Lochness nodes for course-related usage.
  • 1 min read

Office Hours

This spring semester, we are offering drop-in office hours every Monday and Wednesday from 2:00 to 4:00 p.m. Stop by to meet with our student consultants and ask any questions you have about using HPC resources. Whether you’re just getting started or need help with a specific issue, feel free to bring your laptop to walk us through any problems you're facing. There's no need to create a ticket in advance; if follow-up is needed, the student consultants will open a ticket on your behalf, and you'll receive further instructions.

Consulting Hours

  • Date: Every Monday and Wednesday
  • Location: GITC 2404
  • Time: 2:00 PM - 4:00 PM
  • 1 min read

HPC 2025 Spring Events

ARCS HPC invites you to our upcoming events. Please register for the events you plan to attend.

Introduction to Wulver: Getting Started

Save the Date

  • Date: Jan 22nd 2025
  • Location: Virtual
  • Time: 2:30 PM - 3:30 PM

Join us for an informative webinar designed to introduce NJIT's HPC environment, Wulver. This virtual session will provide essential information about the Wulver cluster, how to get an account, and allocation details.

Registration is now closed.

Introduction to Wulver: Accessing System & Running Jobs

Save the Date

  • Date: Jan 29th 2025
  • Location: Virtual
  • Time: 2:30 PM - 3:30 PM

This HPC training event focuses on providing the fundamentals of SLURM (Simple Linux Utility for Resource Management), a workload manager. This virtual session will equip you with the essential skills needed to effectively utilize HPC resources using SLURM.

Registration is now closed.

  • 1 min read

Office Hours

We currently offer drop-in office hours every Wednesday and Friday from 2:00–4:00 p.m. Stop by to meet with our student consultants and ask any questions you have about using HPC resources. Whether you’re just getting started or need help with a specific issue, feel free to bring your laptop to walk us through any problems you're facing. There's no need to create a ticket in advance; if follow-up is needed, the student consultants will open a ticket on your behalf, and you'll receive further instructions.

Consulting Hours

  • Date: Every Wednesday and Friday
  • Location: GITC 2404
  • Time: 2:00 PM - 4:00 PM
  • 1 min read

HPC Fall Events

ARCS HPC invites you to our upcoming events. Please register for the events you plan to attend.

SLURM Batch System Basics

Save the Date

  • Date: Sep 18th, 2024
  • Location: Virtual
  • Time: 2:30 PM - 3:30 PM

Join us for an informative webinar designed to introduce researchers, scientists, and HPC users to the fundamentals of the SLURM (Simple Linux Utility for Resource Management) workload manager. This virtual session will equip you with essential skills to effectively utilize HPC resources through SLURM.

Registration is now closed. Check the HPC training for the webinar recording and slides.

Introduction to Containers on Wulver

Save the Date

  • Date: Oct 16th, 2024
  • Location: Virtual
  • Time: 2:30 PM - 3:30 PM

The HPC training event on using Singularity containers provides participants with a comprehensive introduction to container technology and its advantages in high-performance computing environments. Attendees will learn the fundamentals of Singularity, including installation, basic commands, and workflow, as well as how to create and build containers using definition files and existing Docker images.

Registration is now closed. Check the HPC training for the webinar recording and slides.

Job Arrays and Advanced Submission Techniques for HPC

Save the Date

  • Date: Nov 20th, 2024
  • Location: Virtual
  • Time: 2:30 PM - 3:30 PM

Elevate your High-Performance Computing skills with our advanced SLURM webinar! This session is designed for HPC users who are familiar with basic SLURM commands and are ready to dive into more sophisticated job management techniques.

Registration is now closed. Check the HPC training for the webinar recording and slides.

  • 1 min read

HPC Summer Events

ARCS HPC invites you to our upcoming events. Please register for the events you plan to attend.

NVIDIA Workshop — Fundamentals of Accelerated Data Science

Save the Date

  • Date: July 15, 2024
  • Location: GITC 3700
  • Time: 9 a.m. - 5 p.m.

Learn to use GPU-accelerated resources to analyze data. This is an intermediate level workshop that is intended for those who have some familiarity with Python, especially NumPy and SciPy libraries. See more detail about the workshop here.

Registration

HPC Research Symposium

Save the Date

  • Date: July 16, 2024
  • Location: Student Center Atrium
  • Time: 9 a.m. - 5 p.m.

This past year has been transformative for HPC Research at NJIT. The introduction of our new shared HPC cluster, Wulver, has expanded our computational capacity and made the research into vital areas more accessible to our faculty. Please join us to highlight the work of researchers using the HPC resources and connect with the NJIT HPC community.

Please register for the symposium here, you can also sign up to present your HPC research as a lightning talk or poster presentation:

SLURM Workload Manager Workshop

Save the Date

  • Date: August 13 & 14, 2024
  • Location: GITC 3700
  • Time: 9 a.m. - 5 p.m.

This immersive 2-day experience will take you through comprehensive technical scenarios with lectures, demos, and workshop lab environments. The Slurm trainer will assist in identifying commonalities between previously used resources and schedulers, offering increased understanding and adoption of SLURM job scheduling, resource management, and troubleshooting techniques.

Registration is now closed.

  • 1 min read

Wulver Maintenance

Wulver Monthly Maintenance

Beginning Feb 1, 2024, ARCS HPC will be instituting a monthly maintenance downtime on all HPC systems on the second Tuesday from 9AM - 9PM. Wulver and the associated GPFS storage will be taken out of service for maintenance, repairs, patches and upgrades. During the maintenance downtime, logins will be disabled, users will not have access to their stored data in /project, /home and /scratch. All jobs that do not end before 9AM will be held by the scheduler until the downtime is complete and the systems are returned to service.

We anticipate maintenance to be completed by the scheduled time. However, occasionally the maintenance may be completed earlier than scheduled or could be extended to the following days. A notification will be sent to the user mailing list when the systems are returned to service or the maintenance window is extended. Additionally, users will encounter the cluster service information upon logging in to Wulver during maintenance. Please pay attention to the Message of the Day when logging in, as it will serve as a reminder for upcoming downtimes or other crucial cluster-related information. Users should take into account the maintenance window when scheduling jobs and developing plans to meet various deadlines. Please do not contact the help desk, HPC staff or open SNOW tickets for access to the cluster or data during the maintenance downtime.

  • 1 min read

Lochness Maintenance Updates

Lochness is Back Online!

Lochness is mostly back up after being moved to a new facility. The move required complete disassembly and reassembly of the entire cluster. There are 8 nodes down as they were damaged in the move, repairs forthcoming. Infiniband network issues affect 50 nodes, these are in "drain" state. Currently 120 nodes are fully functional. You can use sinfo to see the exact states of nodes accessible to you. Please email hpc@njit.edu for assistance.