This week All our GPUs are sold at cost price — zero margin on the server. See the cost sheets

This week Every GPU sold at cost price

Getting started

What a server includes

Exactly what comes with a DediGPU server: the hardware, the network, the access, the software, and what is deliberately not included.

3 min read · updated 2 September 2026

A server is a physical machine, built around its cards, that nobody else uses while you rent it. The catalogue lists what each card brings with it; this page says what that means in practice.

Hardware

GPUs
The whole card: no vGPU profile, no time-slicing, no MIG unless you configure it yourself. On SXM nodes the full NVLink fabric between your cards; on eight-card nodes, 3.2 Tbps InfiniBand towards other nodes.
CPU and RAM
Dedicated cores and ECC memory sized per card: a H100 SXM comes with 20 vCPU and 192 GB, times the number of cards.
Storage
Local NVMe, 1.5 TB per H100 SXM card, as a single device. RAID-1 on request when you add extra NVMe.
Out-of-band
An IPMI console with power control and virtual media, reachable through the region's jump host.

Network

Port
10 Gbit/s on workstation and consumer cards, 25 Gbit/s on PCIe data-centre cards, 100 Gbit/s on SXM nodes. Unmetered.
Addresses
One public IPv4 with reverse DNS. Up to eight more at $3 each per month.
Egress
$0. The transit share is a line of the cost sheet; bytes are not metered.
Firewall
None in front of the server. Everything you run is reachable unless you firewall it. Outbound port 25 is closed by default.

Access

Root over SSH from the first minute, with your keys or a generated password. An API token per account. The console for everything else: reboot, rename, term, renewal, cancellation, tickets.

Software

The operating system and template you chose, installed and running before you log in, with the driver and toolkit versions listed on the templates page. A serving template pulls the model you named before first boot.

What is not included, on purpose

  • Backups. We keep none. Your data lives on the server's NVMe and nowhere else.
  • Managed services. We do not administer the operating system. We answer questions about the hardware, the network and the platform.
  • A control plane that can read your data. We see power, link state and a port answering; we do not log in.
  • Priority tiers. Everyone gets the same hardware, the same network and the same support.

Try it on a real server.

One month, at cost, root in under 5 minutes. Support opens with the first server.