Skip to content

Cloud VPS · GPU

G20-160: AMD Instinct MI325X, 256 GB VRAM, 160 GB RAM

G20-160 is 20 vCPU, 160 GB of RAM and 720 GB of NVMe storage, with 1× whole card and 256 GB of VRAM, carrying 15 TB of transfer a month. Root access from the first minute, deployed in the region you pick, and the renewal price is the same figure you order at.

Rendering of a host machine sliced into virtual servers, one highlightedG20-16020 vCPU · 160 GB

Specification

What the machine is.

Full specification of the G20-160 plan
Compute
vCPU20 shared
RAM160 GB
ProcessorShared vCPU — the supplier does not publish the exact platform
VirtualisationKVMYour own kernel, full root, no container shim.
Graphics
CardAMD Instinct MI325X
VRAM256 GB
What you get1× whole cardDrivers, CUDA and the framework stack are yours to install — the images we ship are plain Linux.
Storage
Capacity720 GB NVMe
LayoutNVMe storage on the supplier's platform. We have not verified how it is attached, so we do not describe a layout.
Network
IPv41 included
Transfer15 TB / month (metered)Going past the allowance is not billed automatically — we get in touch about it first.
Notes15 TB of transfer a month. We do not print an uplink figure for this platform, because we have not verified one.
Software
ImagesUbuntu 24.04 LTS, Ubuntu 22.04 LTS, Debian 12, AlmaLinux 9, Rocky Linux 9, Ubuntu 24.04 + DockerEvery image we ship is Linux.
Control panelsNone — we do not resell panel licences
BackupsNo managed backup add-on on this line yet — run your own (restic, borg, rsync) to storage you control
ManagementSelf-managed
Deployment
Regions1 availableDefault region Toronto.
Ready inBuilt to order once the card is confirmed in stock — allow one business day
AvailabilityIn stock

What it costs

$5089.90/month

Renews at $5089.90 a month. The first price is the forever price.


Hourly
$6.973/hr
Annual
$54970.92/yr

10% off twelve months at $5089.90.

Setup fee
None
Billing terms
monthly or annual
Currency
USD

Excludes tax, which is calculated at checkout from where you are and your tax number.

Configure and deploy

Region and image are chosen on the configurator, and the price does not change with either.

Scope

What the price covers, and what it does not.

G20-160 is a self-managed server. The list on the right is not a gap we intend to fill later — it is the other product.

In the price

  • Full root access on a Linux machine with an NVIDIA card attached
  • One dedicated IPv4 for the life of the server
  • The card and its VRAM are stated per size, including where the size is a slice of one GPU rather than a device of its own
  • Provisioned to order once the supplier confirms the card is in stock, and handed over ready to log in

Not in the price

  • Nobody patches the operating system but you — this is a self-managed server
  • NVIDIA drivers, CUDA and the framework stack are yours to install; we ship the plain Linux images and nothing else
  • Provisioning is not instant: GPU stock moves, so the card is confirmed against the order rather than assumed
  • No snapshot, resize or private-networking panel features on this line — it is a different platform underneath
  • No managed backup add-on on this line yet: take your own backups off the box
  • We publish no benchmarks: performance depends on your model, batch size and precision, and a number from us would not survive contact with your workload
  • Prepaid terms are not pro-rated if you destroy the server early

Suited to

Where G20-160 is the right size.

20 vCPU, 160 GB RAM and 720 GB NVMe, 1× whole card, with 15 TB of transfer a month.

Line

G20-160 sits on the GPU line. The card is 1× whole card, with no other tenant on it. The cores beside it are shared, which is rarely the constraint on a GPU machine but is worth knowing before you plan around them.

Questions

About G20-160 specifically.

These are the questions this plan actually gets, answered without the marketing round-off.

What does the renewal cost?
The same 5089.90 you pay today. Our prices do not step up after the first term.
Do I get root?
Yes, from the first minute, on every plan. We do not install agents and we do not watch what runs on your machine.