Place
Head office, or a server room the organisation names.
Hardware
A server chosen for the model, not the other way round.
We ask which model should run, how many people ask at once, and whether a large document index sits beside it. Only then do we propose a machine and deliver it where it has to stand.
Conditions
Power, cooling, noise and rack space are settled before an order.
No tables
We do not publish tokens per second. On your site the number would differ anyway.
What we ask before anything is shipped
- Where the machine will physically stand and who holds the key.
- Whether power and cooling can carry a GPU, not only an office PC.
- Whether server noise is acceptable in that room.
- How many employees will use it at once, and whether the document index runs on the same machine.
- Whether a smaller model does the job without the largest card.
Delivery and bring-up
Hardware is chosen for the task you described and delivered to the place the organisation names. Bring-up covers the system, the model service and administrative access for your people, not a password on a note for everyone.
When the case is small, we say plainly that a GPU is unnecessary. We do not add a card just because the page is called GPU servers.
What you will not find here
There is no price configurator, no card ranking and no promise that a given layout “serves 100 users”. That figure cannot be stated honestly without your model and your questions.
The server is part of a deployment. Hardware alone, without access rules and without people who can restart it, is not yet an assistant.