Magus02 — HPC Knowledge Base

Shiv Nadar Institution of Eminence — HPC Infrastructure
Magus02
The primary High Performance Computing cluster at SNU, providing researchers and students with parallel compute resources for scientific simulation, deep learning, computational chemistry, and data-intensive workflows.
9064
Total Cores
238 TFlops
Peak Performance
57
Compute Nodes
750 TB
Storage

What is HPC?

High Performance Computing (HPC) is the ability to process data and perform complex calculations at high speeds. To put it in perspective: a laptop or desktop with a 3 GHz processor can perform around 3 billion calculations per second. HPC solutions can perform quadrillions of calculations per second — far exceeding anything achievable on a standard workstation.

HPC is specifically needed for the following reasons:

  • It paves the way for new innovations in science, technology, business, and academia.
  • It improves processing speeds, which can be critical for computing operations, applications, and workloads.
  • It helps lay the foundation for a reliable, fast IT infrastructure that can store, process, and analyse massive amounts of data for various applications.

Quick Specifications

TOTAL CORES (AMD) PEAK PERFORMANCE COMPUTE NODES LOGIN NODES MASTER NODES HIGH MEM NODES (SSD)
9064 238 TFlops 57 2 2 12

💡 Memory per Core: Compute Nodes → 4 GB/core (total ~29,184 GB)  |  High Memory Nodes → 16 GB/core (total ~12,288 GB). Use the high_mem queue for memory-bound applications such as Gaussian.

Compute Specification

Node Type Processor RAM Cores GPU No. of Nodes Total RAM Total Cores
Compute Node AMD EPYC 7763 512 GB 128 — 57 29,184 GB 7,296
High Memory Node AMD EPYC 7543 1024 GB 64 — 12 12,288 GB 768
GPU Node Intel Xeon Platinum 8368 512 GB 64 Tesla A100 ×4 2 1,024 GB 128
Totals 71 nodes ~42,496 GB 8,192
Storage 750 TB — Parallel Filesystem

⚠️ High Memory Nodes are intended for applications that require large in-memory datasets but do not scale well across nodes (e.g., Gaussian). Submit to the high_mem queue and use #SBATCH --mem to request appropriate memory.

How to Connect to Magus02

Magus02 supports SSH-based access. Three connection methods are available:

🌐
Web Browser
Visit the login portal directly from any browser on the SNU network.
magus02.snu.edu.in
🐧
Linux / macOS
Use the built-in SSH client from your terminal. Default port is 22.
🪟
Windows (PuTTY)
Download PuTTY and connect using the hostname and port below.
Host: magus02.snu.edu.in
Port: 22
# Connect from Linux / macOS terminal:
$ ssh [email protected]          # default port 22

# Specify port explicitly:
$ ssh -p 22 [email protected]

How to Reset Password on Magus02

01
Log in to your Magus02 account
Connect via SSH using your existing credentials. If you cannot log in, contact the HPC administrator to reset your account.
02
Run the passwd command
Once logged in, use the standard passwd utility. You will be prompted for your current password, then asked to enter and confirm the new one.
[user@magus02 ~]$ passwd
Changing password for user user.
Current password: ••••••••
New password: ••••••••
Retype new password: ••••••••
passwd: all authentication tokens updated successfully.

🔐 Password Policy: Minimum 8 characters including uppercase, lowercase, numbers, and special characters. Shared or weak passwords are strictly prohibited.

How to Connect to Magus from Outside SNU Network

🔒
VPN Required for External Access
To gain access to Magus02 from outside the SNU network, a VPN connection is mandatory. To acquire VPN access, submit a VPN Request Form or scan the QR barcode provided by the HPC administration team. Once connected, SSH works as normal.
# Step 1: Connect to SNU VPN using the provided VPN client
# Step 2: Once VPN is active, SSH as usual:
$ ssh [email protected]

How to Transfer Data using WinSCP (Windows Client)

WinSCP is a free Windows SFTP/SCP client that provides a graphical interface for transferring files between your Windows machine and the cluster.

Step 1 — Open WinSCP and fill in connection details, then click Login
Launch WinSCP. In the Login dialog enter:
  • File protocol: SFTP
  • Host name: magus02.snu.edu.in
  • Port number: 22
  • User name: your username
  • Password: your password
Click the Login button.
Step 2 — Click Yes on the Warning dialog
On first connection, WinSCP will display a host key warning. Click Yes to accept and proceed.
Step 3 — Transfer data using the Key Panel
Once connected, you will see a dual-panel view: local files on the left, remote cluster files on the right.
  • To upload: Right-click a local file → Upload
  • To download: Right-click a remote file → Download
Drag-and-drop is also supported between panels.

💡 Linux / macOS alternative: Use scp or rsync from the terminal.
scp myfile.tar.gz [email protected]:~/

Applications on Magus

The following scientific applications are pre-installed on Magus02. Use the module system to load the appropriate environment before submitting jobs.

🔷 VASP 🔷 Quantum ESPRESSO 🔷 Gaussian
🔷 Siesta 🔷 GROMACS 🔷 OpenFOAM

⚠️ Gaussian users: Gaussian requires high-memory nodes. Always submit Gaussian jobs to the high_mem queue and request sufficient memory with #SBATCH --mem.

Queue Structure on Magus02

Magus02 uses Slurm for job scheduling. Jobs are submitted to specific partitions based on required resources, node type, and runtime. Select the appropriate queue to minimise wait times.

S.No. Processor Architecture Priority Queue Name Min Cores / Job Max Cores / Job Max Nodes Wait Time Max Cores / User Max Pending Jobs Max Running Jobs
1 AMD EPYC 7763 50 medium 32 128 77 1 week 512 2 2
2 AMD EPYC 7043 50 small 8 32 16 1 month 128 2 4
3 AMD EPYC 7543 60 high_mem 32 256 12 2 weeks 1024 2 2
4 AMD EPYC 7763 60 large 64 256 24 2 weeks 1024 2 2
5 Intel Xeon Platinum 8368 — GPU 16 16 4 1 week 32 2 2
6 AMD EPYC 7763 — low / long 32 1024 All nodes 1 month 1024 2 8

✅ Queue selection tip: For small test jobs or debugging → use small. For production multi-node runs → use medium or large. Memory-intensive applications (Gaussian, VASP single-node) → use high_mem. Long-running low-priority jobs → use low/long.

Slurm Command Reference

Common Slurm Commands

# Submit a batch job
$ sbatch job_script.sh

# List all running/pending jobs
$ squeue

# Check only your jobs
$ squeue -u $USER

# Cancel a job by ID
$ scancel JOB_ID

# Check detailed job status
$ scontrol show job JOB_ID

# View partition / queue information
$ sinfo

# Start an interactive session
$ srun --pty --partition=medium --ntasks=4 /bin/bash

Example Slurm Batch Script

#!/bin/bash
#SBATCH --job-name=my_simulation
#SBATCH --partition=medium
#SBATCH --nodes=2
#SBATCH --ntasks-per-node=64
#SBATCH --mem=256G
#SBATCH --time=7-00:00:00          # max 1 week
#SBATCH --output=slurm-%j.out
#SBATCH --error=slurm-%j.err

module load intel/oneapi mpi/intelmpi

mpirun -np $SLURM_NTASKS ./my_program

Slurm Reference PDF

📄 Download Linux / Slurm Command Reference (PDF)

For further assistance, contact the HPC Administration Team or raise a support ticket through the SNU IT helpdesk.

Maintained by HPC Administration · Shiv Nadar Institution of Eminence, Delhi NCR · wiki.snu.edu.in

Retrieved from "https://wiki.snu.edu.in/index.php?title=Magus02_new&oldid=1174"
No categories assignedEdit