Center for High Performance Computing and Big Data

  • WMU CS
  • WMU
  • NSF
  • NVIDIA DVLP
  • CSHPC 2022
  • Home
  • Cluster
  • Research
  • Documentation
  • Classes
  • Events

Cluster hardware description


cluster front

Cluster view

The Thor compute cluster is a high performance parallel computing cluster at WMU. Thor has 22 compute nodes most of which have Nvidia Kepler series K20 GPU accelerator cards. Nodes 2-21 have dual Sandy-Bridge Xeon processors with 128GB of RAM, and a 1 Terabyte hard drive purely for temporary work. Nodes 0 and 1 have four Sandy-Bridge Xeon processors each with a total of 32 processing cores. Node 1 has 256GB of RAM, whereas node 0 has 512GB.

Binding the entire cluster together is an ethernet management network and a low-latency Infiniband communication network. All nodes have access to network mounted directories on a high performance, high redundancy RAID file server. For user tasks, a main head node is used to manage the entire cluster and share programs and data.

Cluster detailed information

ID# Description Qty

Hardware
1 Relion 2800 Head Node
  • Dual 1100W Power Supply
  • Dual Intel Xeon E5-2670,8C,2.6GHz,1600MHZ Ram, 115W
  • 128GB, DDR3-1600 ECC(16 x 8GB)
  • HW Raid 0/1/5/6/10/50:LSI/3ware9750-8i w/ cache and BBU
  • RAID 1 Volume: 476 GB (2+0 x 500GB SATA2)
  • RAID 5 Volume: 7629 GB (5+1 x 2TB SATA2)
  • HCA, QLogic QLE7340, 1-port QSFP QDRmemfree
  • DVD-ROM
  • Preload, CentOS, Version 6
1
2 Relion 2800GT Kepler GPU Node - 128GB
  • Dual 1620W Power Supplies
  • Dual Intel Xeon E5-2670, 8C, 2.6GHz, 1600MHz RAM, 115W
  • 128GB, DDR3-1600 ECC (16 x 8GB)
  • 1TB, SATA2 7200rpm
  • Integrated SATA Controller, 2x 6Gbps, 6x 3Gbps ports
  • Integrated InfiniBand QDR
  • Preload, CentOS, Version 6
19
3 Relion 2800 GT Fermi GPU Node - 128G B
  • Dual 1620W Power Supplies
  • Dual Intel Xeon E5-2670, 8C, 2.6GHz,1600MHz RAM, 115
  • 128GB, DDR3-1600 ECC (16 x 8GB)
  • 1TB, SATA2 7200rpm
  • Integrated SATA Controller, 2x 6Gbps, 6x
  • 3Gbps ports
  • Integrated InfiniBand QDR
  • NVIDIA Tesla M2090
  • Preload, CentOS, Version 6
1
4 Relion 2804e Compute Node - 32 Core, 256GB
  • Dual 1600W PS
  • Quad Intel Xeon E5-4640, 8C, 2.4GHz,
  • 1600MHz RAM, 95W
  • 256GB, DDR3-1600 ECC (16 x 16GB)
  • 1TB, SATA2 7200rpm
  • Integrated SATA Controller
  • HCA, Mellanox ConnectX 3, 1x QSFP/QDR
  • IB
  • DVD-ROM
  • Preload, CentOS, Version 6
1
5 Relion 2804e Compute Node - 32 Core, 512GB
  • Dual 1600W PS
  • Quad Intel Xeon E5-4640, 8C, 2.4GHz,
  • 1600MHz RAM, 95W
  • 512GB, DDR3-1600 ECC (32 x 16GB)
  • 1TB, SATA2 7200rpm
  • Integrated SATA Controller
  • HCA, Mellanox ConnectX 3, 1x QSFP/QDR IB
  • DVD-ROM
  • Preload, CentOS, Version 6
1
6 IceBreaker 4836 Storage Server - 72TB
  • Redundant 1400W Power Supplies
  • Dual Intel Xeon E5-2603, 4C, 1.8GHz,
  • 1066MHz RAM, 80W
  • 32GB, DDR3-1600 ECC (8 x 4GB)
  • HW RAID 0/1/5/6/10/50: LSI 9750-8i 6Gbps
  • SATA/SAS Controller w/ BBU
  • HW RAID 0/1/5/6/10/50: LSI 9750-8e
  • 6Gbps SATA/SAS Controller w/ BBU
  • RAID 1 Volume: 57 GB (2+0 x 60GB MLC SSD)
  • RAID 6 Volume: 17166 GB (8+1 x 3TB SATA2)
  • RAID 6 Volume: 17166 GB (8+1 x 3TB SATA2)
  • RAID 6 Volume: 20027 GB (9+0 x 3TB SATA2)
  • RAID 6 Volume: 17166 GB (8+0 x 3TB SATA2)
  • HCA, QLogic QLE7340, 1-port QSFP QDR memfree
1

RACK Infrastructure
7 Cluster Kit, Rack, Basic (Includes 42U rack,
  • crate and blanks)
2
8 PDU, Metered, 0U, 12.5kW, 3ph, 208V,
  • CS8365C/(30)C13&(6)C19, 10'
4
9 PDU, Basic, 1U, 20A, 120V, 5-20P/(10)5-20,
  • 12' (For UPS)
2
10 UPS, Smart-UPS 3000VA USB LCD RM 2U 120V 1
11 UPS, APC Network Management Card 2 1
12 Rackmount 17&Quot; LCD/ US Keyboard/ USB
TouchPad, 1U, Avocent
1

Ethernet Networking
13 Switch, Netgear, 48-port, GigE (Cluster
Management Switch)
1
14 Switch, Netgear, 48-port, 10/100 (IPMI
Switch)
1

Infiniband Fabric
15 Switch, QLogic 12300, 36 x QDR/QSFP,
  • managed, 2 PSU, 1U, w/ rails
1
16 Switch, QLogic 12300, Fabric Suite Software 1
17 Switch, QLogic 12200/12300, Fabric Suite SW 2
18 Cable, InfiniBand, Passive Copper,
QSFP/QSFP, 30AWG, 2m
15
19 Cable, InfiniBand, Passive Copper,
  • QSFP/QSFP, 28AWG, 3m
7
` Software
20 Scyld Cluster Ware HPC6.0 (License)
  • Scyld ClusterWare Base Cluster License
  • inc. 1 Master Nodes
1
21 3 years support Scyld Cluster Ware HPC 6.0 1
22 License Scyld Insight, 4 Concurrent Sessions
1
23 1 Year Support, Scyld Insight
3
24 PGI PGI Accelerator CDK 16 processes
  • License 2 User (part # 018AL-C2-16-A)
1
25 PGI PGI Accelerator CDK 16 processes 3
  • Nvidia