Hardware - chunxxc/GPU-Server-Handbook GitHub Wiki

In this section, we detail the hardware available at each of the servers. This information can be useful to predict how much speed-up you can expect when transitioning from your local computer to the servers, or from one server to the other.

Most of the information included here is accessible from your command-line access, where launching nvidia-smi, lscpu, lshw, cat /etc/fstab, or cat /proc/partitions will detail the included hardware and some of its specifications.

DGX-station server (ise-dgx.ee.kth.se)

Nvidia DGX-station

  • Number of V100: 4
  • VRAM Each: 32G
  • Number of CPU: 40
  • RAM Total: 252G

System SSD drive

  • 1.8T mount on /

External SSD drive

  • 3X1.8T raid0 array mount on /raid

GPU-equipped server 1 (gpu1.ee.kth.se)

Nvidia Tesla P100-PCIE-16GB

SSD

The partition table for the SSD is arranged as follows, loaded in /dev/sda.

Number Start End Size File system Mount-point
1 1049kB 538MB 537MB fat32 /boot
2 538MB 108GB 107GB ext4 /
3 108GB 967GB 859GB ext4 /data
4 967GB 1000GB 33.3GB swap none
Partition number 1 is the boot partition (mounted in /boot/). Partition number 2 is the system partition, where the system-wide installed programs are located. Partition number 3 is the fast data partition, where each user has a directory /data/*username* to place data or personal programs that will need to be loaded promptly to speed up experiments. This fast data directory is linked to each user's home directory as /home/*username*/fast_data. Partition 4 is SWAP space to ensure the system does not crash if the system runs out of RAM.

HDD

The hard-disk drive /dev/sdb of 3.6T only has one partition and is mounted in /home, containing most of the user files.