About

CBL Compute Server

Goal: Setup a compute server for use by the lab with a large amount of CUDA processing power.
Solution: A linux based server with job scheduling and various software packages installed.

Software Installed

  • Torch
  • Caffe
  • Slurm
  • Theano
  • OpenCV
  • TensorFlow

Hardware

Specifications

  • CPU: 2 x Intel Xeon E5-2609 V3
  • RAM: 128GB
  • GPU: 4 x Nvidia Titan X (12GB VRam)
  • Networking: 10Gb Fiber
  • HDD: 4 x 1TB disks in RAID 5

Card Installation

Two GPUs were installed for each CPU. Besides those, there was also a RAID card installed for the front HDD backplane, and a 10Gb fiber card for high speed data transfer.

Networking

The server was setup on a separated network from the broader university network. Only specific network ports could connect to it, preventing public Wi-Fi users from accessing it and making the system more secure.