High Performance Computing Clusters

The Divison of IT is working on launching a new High Performance Computing Cluster for the Stevens community. Below you will find the specifications of the cluster: 

• Condominium model cluster  

• 46 Compute Nodes 

• 3520 Cores 

• 8 NVidia H100 GPUs 

• 53TB RAM, avg. ~16GB/core 

• 1PB of storage, with an additional 1PB for data backups 

• Will have multiple queues (Standard, high memory, and GPU queues) 

• Will have high-speed, low-latency InfiniBand Networking 

• Dedicated high-speed storage  

DuckUte is a mixed High Performance Computing cluster at Stevens, which extends our compute capabilities for machine learning, simulation and modeling, and a wide range of other applications. This HPC cluster, formerly known as oHPC, is built with HPE BladeSystem-based servers and has a unique combination of the CPU and GPU compute nodes, managed by the SLURM scheduler. 

• OpenHPC stack backed by the Linux system 

• 31 shared mixed CPU/GPU compute nodes (8,10 or 12-core Intel® CPU E5-2680 251GB RAM or NVIDIA Quadro K3100M 4GB RAM) 

• Total number of CPUs: 1152 

• Total number of CUDA cores: 10,752 

• Up to 140TB Storage (HPE 3PAR StoreServ Storage Systems) 

• 10Gb interconnects 

• SLURM-based resource manager and job scheduler 

• Standard development tools and custom-built applications 

Available to any researchers on campus (includes faculty, staff, and sponsored students).  Software Available on DuckUte 

• C, C++, Python, R, Matlab, Silo, Valgrind 

• Additional software can be requested 

Additional resources will be added at the end of 2023 and throughout 2024. 

A secure and isolated environment is available for researchers with specific compliance requirements such as controlled unclassified information CUI, DOD and other government compliances needs.   

If you are interested in learning more about the CUI Environment, contact our team at [email protected].  

Dorothy will be replaced/upgrade by the new HPC cluster. 

Dorothy, our traditional general-purpose high-performance computing (HPC) system, available on request to campus researchers. 

• 80 compute nodes  

• 900 cores  

• 4GB+ RAM/core  

• 50TB of Storage  

• C, C++, Fortran, Python, Matlab  

• 1Gb Networking  

• Slurm job scheduler 

Pharos, the private high-throughput and data-intensive computing cluster used exclusively by Davidson Labs and the CEOE department.  

• 64 compute nodes  

• 1280 cores  

• 6GB+ RAM/core  

• 2 PB of Lustre distributed storage  

• 56 Gbps Mellanox Infiniband fabric for compute nodes and storage system  

• C, C++, Fortran, R, Python, Matlab  

• Moab/Torque job scheduler 

Contact Dr. Raju V. Datla for more information on Pharos.