Biowulf High Performance Computing at the NIH
History of the NIH Biowulf Cluster

[Current hardware] [HPC staff] [Careers] [Policies] [Contact]

1999 NIH Biowulf Cluster started with 40 "boxes on shelves".

CHARMm and Blast running on cluster.

14 active users

2000 1st scientific paper citing Biowulf

First batch of 16 Myrinet nodes added to cluster.

Swarm developed in-house to submit large numbers of independent jobs to the cluster.

2001 Blat, GAMESS, Gauss, Amber, R available on cluster.

PBS Pro batch system.

2002 New login node

80 nodes added to cluster.

Mysql databases available to users

NAMD installed

2003 New Biowulf website

Added 198 'p2800' nodes, including 24 nodes with 4 GB of memory

Default user disk quota increased from 32 to 48 GB

Myrinet upgrade

Smallest memory node on Biowulf has 512 MB RAM

First benchmarking of 64-bit Opteron processor

66 2*2.8 GHz Intel Xeon, 2GB RAM nodes added to cluster

2004 Added 132 dual-processor nodes 2.8 GHz Intel Xeon with 2-4 GB memory, + 32 AMD Opterons with Myrinet, + 42 AMD Opterons.

Cluster reaches 1000+ nodes.

Swarm bundling option for 1000s of short jobs

Adios Telnet!

2005 Added 64 dual-processor AMD Opterons, 2.2. GHz with 4 GB RAM.

Nodes upgraded to RHEL 3.1 (Linux 2.6 kernel)

New login node, dual-processor 3.2 GHz Xeon, 4 GB memory

Quad-core 2.2 GHz 64-bit AMD Opteron for interactive jobs

100th scientific paper published citing Biowulf

2006 Added 324 dual-processor, 2.8 GHz AMD Opterons with 4 GB memory.

Added 40 dual-core, 2.6 GHz AMD Opterons with 8 GB memory.

Added 64 Infiniband-connected nodes.

Firebolt, an SGI Altix 350 with 32 Itanium-2 processors and 96 GB of memory, is installed as a 'fat node' on Biowulf

Biowulf total: 3700 CPUs

2007 Another 48 Infiniband-connected nodes for a total of 112 IB nodes

Per-user limit raised from 16 to 24 IB nodes.

280 2.8 GHz Opterons, 4-8 GB RAM added to cluster

2008 Helix transitions from SGI Origin 2400 to a Sun Opteron

All nodes go to Centos 5

2009 "The NIH Biowulf Cluster: 10 years of Scientific Supercomputing" symposium held at NIH, Bethesda..

Added 224 Infinband-connected nodes, 8 processor, Intel EMT-64 2.8 GHz.

Added 16 "Very-large memory nodes", 72 GB memory each + one 512 GB memory node

Storage system added which increases capacity by 450 TB.

2010 Myrinet nodes decommissioned

All Biowulf nodes now 64-bit

Added 336 Intel quad-core Nehalem nodes, 2.67 GHz, 24 GB memory + 16 nodes with 72 GB RAM

16 Pilot GPU nodes added to cluster.

Biowulf has a total of 9000 cores.

2011 Helix transitioned to 64-processor, 1 TB memory hardware.

Added 328 compute nodes, 2* Intel 2.8 GHz X56660 processors, 24 GB memory.

500th scientiific paper published citing Biowulf

2013 Annual Biowulf account renewals implemented

Environment modules implemented for scientific applications

2014 NCI and NIMH fund nodes on the cluster

Major network upgrades increasing bandwidth between HPC systems and NIH core

24 NIH ICs, 250 PIs, 620 users

2015 Storage reaches 3 PB

NCI and NIDDK fund additional nodes

1000th scientific paper published citing Biowulf

Phase 1 HPC upgrade installed

Transition to Slurm batch system

Biowulf2 goes production in July

Webservers migrated to new hardware

Dedicated data-transfer nodes installed

400+ scientific applications installed and updated

2016 New 'HPC account' process: all users get access to both Helix and Biowulf

Monthly Walk-in Consults started

30,000 cores added

3 PB disk storage added

All new nodes on FDR Infiniband

Singularity containers on Biowulf

1500th paper published based on Biowulf usage

Biowulf is #156 on the Nov 2016 Top500 Supercomputer list

2017 Phase 3 goes online, with:
    30,000 additional cores
    72 2*K80 GPU nodes
    1.5 TB memory nodes
    3 TB memory nodes
    All new nodes on EDR Infiniband

HPC staff develops new classes on Object storage, Singularity and Snakemake

NIH Director's Award for HPC Expansion Team

Biowulf is #66 on the June 2017 Top500 Supercomputer list

2018 Biowulf Seminar Series

48 P100 nVidia GPU nodes added to cluster, each with 4 GPUs

2000th paper published based on Biowulf usage

Biowulf cluster migrated to Centos7

Four V100 nVidia GPU nodes added to cluster, each with 4 GPUs

'Intro to Biowulf' becomes an online class

Helix becomes the Interactive Data Transfer system, and moves to newer faster hardware.

2019 Biowulf 20th anniversary Seminar Series

Additional SSDs added allowing up to 2.4 TB of local scratch disk allocations

2500th paper published based on Biowulf usage

New classes developed and taught:
Scientific Python for Matlab Users
Deep Learning by Example on Biowulf
Managing Personal Software Installations

2020 21000+ CPUs added to Biowulf, including:
56 GPU nodes with Nvidia V100 GPUs, 4 per node, Nvlink
Skylake processors, 384 GB memory and 3.2/1.6 TB SSDs

3000th paper published based on Biowulf usage

Biowulf supports COVID-19 research

Monthly Zoom-In Consults started

2021 4000th paper published based on Biowulf usage

Interactive Krona-based visualizations of data directories made available to users

New visualization partition with GPU-accelerated graphics

2022 9000 additional Epyc CPUs added to cluster

88 A100 GPUs (22 GPU nodes) added to Biowulf