Zest - High-Performance Computing

Cluster Type: High-Performance Computing (HPC) Scheduler: Slurm

What is Zest?

Zest is Syracuse University’s high-performance computing cluster with over 25,000 cores, designed for research work that requires tying together multiple compute nodes or cannot be split into smaller components. Compute nodes are interconnected with InfiniBand to pass information between nodes with much lower latency than Ethernet.

Best For:

  • Multi-node parallel jobs (MPI)
  • Tightly-coupled computations requiring fast interconnect
  • Large-scale parallel workloads (hundreds of cores)
  • GPU-accelerated research
  • Long-running jobs (up to 40 days)
  • High-memory workloads

Technical Overview

Scale: Over 25,000 cores with InfiniBand interconnect

Architecture:

  • Traditional HPC cluster with uniform compute nodes
  • InfiniBand fabric for low-latency node-to-node communication
  • Optimized for MPI and parallel workloads
  • Multiple partitions for different workload types

Job Capabilities:

  • Multi-node jobs (scale across dozens or hundreds of nodes)
  • MPI parallel computing with fast interconnect
  • High core count (hundreds of cores per job)
  • High-memory configurations available
  • Extended runtimes (20-40 days depending on partition)

GPU Capabilities:

  • GPU nodes available through SUrge infrastructure
  • Primarily NVIDIA A40 GPUs
  • Request with --gres=gpu:1 in SBATCH scripts
  • GPU partitions available for accelerated computing

Storage: 4TB home directory (default), NetApp-based, automatically mounted on all compute nodes

Access: SSH via its-zest-loginX.syr.edu

Partitions

Partition Purpose Max Runtime
normal (default) CPU-intensive workloads 20 days
compute_zone2 CPU-intensive workloads 20 days
longjobs Extended runtime needs 40 days
gpu, gpu_zone2 GPU computations 20 days

How It Works

Submit jobs using Slurm batch scripts:

  1. Create your SBATCH script with resource requests
  2. Submit with sbatch script.sh
  3. Slurm schedules based on availability and priority
  4. Job runs on allocated compute nodes
  5. Results saved to your home directory

Typical Use Cases

  • Molecular dynamics simulations (GROMACS, NAMD)
  • Deep learning model training with GPUs
  • Computational fluid dynamics
  • Multi-node parallel computations (MPI)
  • Weather/climate modeling
  • Finite element analysis

Learn More

📊 Zest Technical Specifications
💻 Zest Code Examples

Table of contents