Virtual cluster

High-performance computing system specifications

The University cluster is a set of high-performance processors and software used for conducting complex scientific calculations for research that require significant power and data storage.
The virtual cluster has a total of 23 nodes, one of which is a front-end node, with the following characteristics:

Node 1 (front-end)
  • 8 AMD EPYC 7413 cores @2.65GHz
  • 32 GB RAM
  • 300 GB local storage
  • 30 TB of NAS storage for home
  • Ubuntu 24.04 LTS operating system
Nodes 2-3
  • 20 Intel(R) Xeon(R) Gold 6148 CPU cores @ 2.40GHz
  • 200 GB RAM
  • 300 GB local storage
  • 30 TB NAS storage for home
  • Ubuntu 24.04 LTS operating system
Nodes 4-23
  • 20 AMD EPYC 7413 cores @2.65GHz
  • 200 GB RAM
  • 300 GB local storage
  • 30 TB of NAS storage for home
  • Ubuntu 24.04 LTS operating system

In addition, there are a number of GPUs connected to specific nodes:

  • 2 NVIDIA Tesla GPUs connected to the node 2
  • 1 NVIDIA A40 GPUs connected to the node 3

The cluster includes, among others, the following software and tools:

  • Matlab
  • Stata
  • R
  • Gromacs 2024
  • Python
  • NVIDIA CUDA Toolkit
  • other standard tools already in the OS

To ensure that all active projects have access to resources, each job must run for a maximum of 24 hours. For jobs that necessarily require more execution time, a specific queue with a maximum duration of 10 days is also provided.
Jobs must be scheduled via Slurm; exceptions are Matlab and Stata tasks, which run on software not handled by the job manager.
There is a dedicated queue on Slurm for testing and debugging tasks, which is limited to 1 hour.

Request for cluster access and ‘Research’ VPN authorisation 

To access the cluster, you must first ensure you have the "Ricerca" [Research] profile active for the University VPN; if you do not have it yet, you will need to open a ticket to request it. Additionally, you must request explicit access to the computing cluster itself. Both requests can be submitted by opening a ticket under the queue "Assistenza e guasti > Supporto ricerca > Cluster di calcolo". To find out who should submit the ticket and what information is required, please visit www.unive.it/vpn-eng.

Instructions for use

Access

  • Connect via VPN
  • Select the "Ricerca" profile
  • Connect to the frontend node:
  • Protocol: SSH
    • user: Your unive account
    • password: Your unive account password
    • server: vhpc01.vhpc.unive.it
    e.g.: ssh user.name@vhpc01.vhpc.unive.it
  • Job execution: Only using the Slurm scheduler.

Storage

You have a storage area available in your personal home directory which cannot exceed 2TB. For data exceeding 2TB, and generally as a good practice for data that isn't immediately needed or for partial output from job executions, we ask that you use your dedicated folder in /scratch/${USER}_no_bkup. The only difference is that the second location is not subject to backup.

Scheduler

  • Jobs on the cluster must be run using the Slurm scheduler.
  • Refer to the official Slurm archive for version-specific manuals.
  • Run sinfo --version in your terminal to check which Slurm version the cluster is currently running.
#!/bin/bash
#SBATCH --job-name=job_name
#SBATCH --comment="job_comment"
#SBATCH --output=/home/user/job_folder/%x_%j.out
#SBATCH --error=/home/user/job_folder/%x_%j.err
#SBATCH --mail-user=user@domain,user2@domain
#SBATCH --mail-type=ALL
#SBATCH --time=24:00:00
#SBATCH --mem=16gb
#SBATCH --nodes=1
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=8

RUNPATH=/home/user/job_folder
cd $RUNPATH
source .venv/bin/activate
export PYTHONPATH=$PYTHONPATH:/home/user/job_folder
srun python -u run_script.py <options>

Queues and partitions

The cluster has 22 compute nodes, a frontend node (the only one you can connect to), and the nodes are organized into different partitions or queues. The list of active partitions on the cluster can be viewed with the "sinfo" command. The maximum duration for the "longjob" queue is 10 days, while the duration for the other partitions is 24 hours.
Example of how to ensure your job restarts after 24 hours.

GPUs

The cluster is equipped with three GPUs: 2 NVIDIA Tesla GPUs on node2, and 1 NVIDIA A40 GPU on node3. To request and use a GPU correctly in your sbatch script, you must include the following directives:
#SBATCH --partition=gpus
#SBATCH --gres=gpu:a40:1 (or #SBATCH --gres=gpu:tesla:2)
For an up-to-date list of available GPUs, please refer to the output of the command sinfo -o "%.12G %.14N" | grep -v null. Please ensure your code follows the official Slurm documentation and utilizes the appropriate libraries (e.g., CUDA, cuDNN) required for GPU acceleration. We offer an additional queue equipped with 3 NVIDIA A40 GPUs, designed for jobs with a maximum runtime of 10 days and a limit of 2 CPUs (e.g., dedicated inference workloads). To gain access to this queue, please submit a request ticket to ASIT via our support portal under "Assistenza e guasti > Supporto ricerca > Cluster di calcolo".

Maintenance

As indicated at the bottom of page www.unive.it/ict-support-for-research, remember that scheduled monthly maintenance of the computing cluster will be performed every first Monday of the month. Cluster nodes will be set to drain starting from the preceding Wednesday to minimize active jobs and will be reactivated at the end of maintenance. Running jobs with appropriate checkpoints will be preserved.

Additional information

Additional information is provided in the welcome message when accessing the frontend node.
For any requests or issues, please submit a ticket (if you’re using a student account, your Supervisor will need to submit it on your behalf) to ASIT via the dedicated queue "Assistenza e guasti > Supporto ricerca > Cluster di calcolo".

Last update: 25/06/2026