First GPU SuperServer for DOCC

Created on February 01, 2024

2024   ·   ai   hardware   research

Dell HGX GPU SuperServer

Early 2024, we ordered our first GPU SuperServer for the Data Observation and Computation Collaboratory (DOCC) project. This server will be used to run AI/ML applications, data analytics, simulations, and visualization tasks. It contains 4 Nvidia H100 in SXM form-factor. It was delivered this Frebruary. The specifications of the server are as follows:

  • PowerEdge XE8640
  • Intel Xeon Platinum 8468 2.1G, 48C/96T, 16GT/s, 105M Cache, Turbo,
    • HT (350W) DDR5-4800
  • 64GB RDIMM * 16: 1TB RAM
  • 3.84TB Enterprise NVMe SSD
  • BOSS-N1 controller card +with 2M2. 480GB (RAID 1)
  • Fully Redundant 3+1 (or 2+2 FTR), Hot-Plug PSU, 2800W
    • M HLAC (200-240Vac) Titanium, C2 Connector
  • NVIDIA HGX H100 4-GPU SXM 80GB HBM3 700W GPU Assembly
  • Broadcom 57414 Dual Port 10/25GbE SFP28 Adapter, PCle Ful Height, V2
  • Broadcom 5720 Dual Port 1GbE Optional LOM
  • Nvidia ConnectX-6 Dual Port 10 GbE QSFP56 Adapter, OCP 3.0