Dedicated Instant Custom GPU Unmetered Domains Pricing Data Centers Migrate
Configure server
Ryzen AI Dedicated GPU

Do you need the graphics card?

A Ryzen AI machine serves a small model from its own system memory. A dedicated GPU earns its price at a few well-defined moments. This page is the dividing line between the two, stated plainly.

The machines without one Every card, priced

What the small machine already covers

Before pricing a card, it is worth being clear what a whole Ryzen AI machine does without one — because for a lot of inference, this list is the whole job.

  • A small quantised model served from system memory
  • One tenant on the metal, full root access
  • Your own runtime, pinned to your own versions
  • NVMe storage for corpora and checkpoints
  • A gigabit port and an IPv4 address included

The model outgrew the memory

Weights that no longer fit do not run slowly — they do not load. Past what system memory holds, the answer is a card chosen by memory first, and how much a model needs is worked through on the LLM page.

The users multiplied

Serving many people at once is what a card’s memory bandwidth and batching are for. That is the point at which a queue on a small machine turns into a case for a dedicated card.

You started fine-tuning

Training holds the weights, the gradients and the optimiser state in memory at once — several times what serving needs. That is card territory, and sometimes more than one card.

Milliseconds became the product

When reply pace is what your customer experiences, the card’s faster memory is the honest fix. A small machine is for jobs that think in seconds.

FAQ

How do the Ryzen AI machines compare to GPU servers?
Same job, different memory. A Ryzen AI machine holds the model in system memory — roomy for the price, slower to read — and a GPU server holds it in card memory, scarcer and much faster. The model size and the traffic pick between them; the badge on the box does not.
Can you help me move a workload across?
Yes. Bring containers, models and data, and ask our engineers on chat before you order — moving between the two shapes is routine in both directions, and easier to plan before the first invoice than after it.
Why are there no prices on this page?
Because they live where they cannot drift: the Ryzen AI machines are priced live on their own page, and every card we fit is priced on the GPU page against the exact machine it goes in. A figure repeated here would only be a second copy waiting to go stale.