Compute Systems – Pascal, Penguins, and Family

“Pascal.valpo.edu”

In 2017 NVIDIA’s Educator GPU Grant program made specialized hardware available to several groups at the University. Two Titan X Pascals were provided. One was given to CIS in response to a proposal submitted by CIS faculty and Professor Jon Beagley of MST. The other was sent to Electrical and Computer Engineering and was later generously transferred to CIS. These were combined with assets provided by the University’s central IT team and Steve Vittitoe, a local University benefactor. This set of components allowed students and staff to build a new specialized server named pascal.valpo.edu.

It provides MST/CIS faculty and other collaborators with a locally administered asset for complex numerical work.

Specifications

Computation

  • Two NVIDIA Titan X Pascal GPUs (~7000 CUDA cores)
  • Two Intel – Xeon E5-2620 v4 x86-64 V4 2.1GHz 8-Core Processors (16 physical cores)
  • added: NVIDIA SLI Bridge unit to link the GPU cards [‘4 slot’]
  • Motherboard – Supermicro MBD-X10DAX EATX Dual-CPU LGA2011-3

Storage

  • 120 GB SSD (“/” – OS/application storage)
  • 120 GB SSD (/media/ssdstorage – ~70 GB for additional application storage; plus the swap partition – ~39 GB )
  • 500 GB SSD (/media/ssdstorage2 – high speed user storage)pending.
  • 3 TB 7200 RPM Standard HDD (/media/storage)
  • 3 TB Variable-speed Standard HDD [Western Digital ‘green’] (/media/storage2)
  • 3 TB Variable-speed Standard HDD [Western Digital ‘green’] (/media/storage3)

Memory

  • System: 128 GB Registered DDR4-2133 SDRAM (upgrade from 64 GB initial configuration on 7/10/2018)
  • Titan X Pascal:12 GB RAM per card (24 GB total)

Power

  • EVGA SuperNOVA G2 1300W 80+ Gold Fully-Modular ATX PSU, attached to an
  • external UPS unit – a CyberPower Systems LX1500GU (1500VA AVR)

Cooling

  • Corsair – H100i Liquid CPU Cooler
  • Corsair – H80i Liquid CPU Cooler
  • Large case and power supply fans

Current software:

  • OS: GNU/Linux Ubuntu 18.04 (upgraded Fall 2019; originally GNU/Linux Ubuntu 16.04.3 LTS)
  • Libraries: CUDA suite(s) v9.2, 10.0, 10.1 (/usr/local/CUDA) [Driver Version: 440]
  • Languages: Oracle/Sun Java 11, GNU Fortran, GNU C, GNU C++, Python v2 (python); Python v3 (python3)

Pascal has access to /mnt/sharedFiles over a standard network connection, see below.

[Particular thanks to Charles Morris, the first principal student sysadmin, who helped build the system.]

“The Mathematicians”

A set of additional machines was brought online in stages beginning in Spring 2019, all running GNU/Linux Ubuntu 18.04 or 20.04 LTS. These are:

  • euler.valpo.edu (AMD Ryzen 5 1600X @ 3.60 GHz / 32 GB RAM; GEFORCE GTX 1060)
  • germain.valpo.edu (Two Intel Xeons E5-2667 @ 2.9 GHz – 12 cores total / 48 GB ECC RAM; GEFORCE GTX 1060) *
  • katherine.valpo.edu (Two Intel Xeons E5-2630 @ 2.30GHzl – 12 cores total / 48 GB ECC RAM; Quadro K2000)
  • fermat.valpo.edu (Intel Core i3-6100 2-Core @ 3.70GHz /  20 GB RAM; GEFORCE GTX 1060)
  • noether.valpo.edu (Intel Celeron G3930 2-Core @ 2.90GHz / 16 GB RAM; GEFORCE GTX 1060)
  • mandelbrot.valpo.edu (Intel Core i5-4570 4-Core @ 3.20GHz / 32 GB RAM; GEFORCE GTX 660)

A private network links the systems via secondary ethernet cards.  An NFS mount (/mnt/sharedFiles) is provided by noether to the other systems over that private network.

[* thanks to Aaron Adams ’19, who donated both “germain” and his time to make it fully operational.]

“The Penguins”

Further donations made additional systems available; these were arranged into their own cluster, with each node named for a species of penguin:

  • Macaroni2 (AMD Phenom II X6 1100T @ 3.7 GHz / 16 GB RAM; 500 GB SSD) **
  • Rockhopper (-/32 GB RAM; 256 GB SSD) ***
  • Chinstrap (pending, -/32 GB RAM; 256 GB SSD) ***

[with thanks to Karl Schmitt(**) and Valpo alumnus Ted Henderson(***) for their contributions to these systems – and Alex Kaariainen and Craig Behnke for help on the builds.]

Other Systems

Several other systems can be used as auxiliary compute nodes, assuming they are not supporting courses as front ends to the small form factor parallel nodes:

  • plaid.valpo.edu
  • tweed.valpo.edu

These are not linked to the private network, but do have access to the same NFS shares via standard University connections. Two systems also play supporting roles, and are linked to the private network:

  • ferris.valpo.edu (Intel Celeron J1900 Celeron on ASRock IMB-151D Bay Trail / 8 GB RAM) – similar to the LittleFe nodes – and
  • a Synology DS418 network attached storage (NAS) system (four 4 TB RAID 10 disks providing ~8 TB storage) providing local-control large storage to support both Windows and Linux systems

Additional Software

See the Cluster Use page for more info.

[Thanks also to Dane Towner and Erik Kispert for many assists with system build-out and various software help.].