Flaem

GPU servers and AI models sold from your own panel

Companies that train models, run them or render graphics need graphics cards, not just CPUs. Flaem lets you sell GPU servers the way you sell dedicated servers and VPS: from a machine inventory, with telemetry for every card, with invoices and renewals. On the same cards you run AI models and sell access to them.

Server with graphics cards
Card 1
Load
Card memory
Temperature
Power draw
Card 2
Load
Card memory
Temperature
Power draw
Card 3
Load
Card memory
Temperature
Power draw
Card 4
Load
Card memory
Temperature
Power draw
  • Working for a clientA card in a dedicated server or in a client virtual machine.
  • Running an AI modelThe card stays with you and answers requests to the model.
Every card has its own telemetry, whether it works for a client or for a model.
01What you sell

Three ways to sell graphics card power

The same card can go to a client with a whole machine, with a virtual machine, or stay with you and compute AI model answers. You choose which of these offers you switch on.

02Card telemetry

Telemetry for every card: load, memory, temperature and power draw

With a GPU, the first question is whether the card is actually working and whether it is throttling. The panel shows it for each card separately, next to the CPU, memory, disks and network of the whole machine.

Load
How busy the card is with computation.
Card memory
How much card memory the model or job takes before it runs out.
Temperature
Whether cooling keeps up under long, full load.
Power draw
How many watts the card pulls, which shows whether it runs at full speed.
GPU servers in the operator panel: load, memory, temperature and power draw of every card, plus the models offered to clients with their keys

Card load over time

lowhigh
Card 1
Card 2
Card 3
Card 4
now

Plus the whole machine

  • Controller sensors: temperatures, fans and power draw
  • CPU, memory and disks
  • Network traffic with real peaks
  • Percentiles p50, p90, p95 and p99
  • Events on the timeline: reboot, reinstall, package change
03AI models

AI models as a service, on your cards and under your brand

Not every client wants to rent a whole server with a card. Some only want to send a request to a model and get an answer. You sell that access as a package, and the model computes on your hardware.

Clients with a package
access keyaccess keyaccess key
AI model
Your cards
Your support
ticket translationfirst line
  1. 01

    You choose the models

    You decide which models you run and on which cards they work.

  2. 02

    The client buys a package

    The package has its price and billing period, like any service in your store.

  3. 03

    The client gets access

    The model address and key are in their panel, with no server to manage.

  4. 04

    The model runs with you

    Requests and answers never leave your infrastructure.

The same model in your support

A local model translates tickets from clients who write in another language and answers on the first line before a case reaches a person.

04Billing

Bill GPU servers like any other service

A server with a card is expensive, so the contract has to be clear from day one. You set the period, price, trial and minimum term in the package, and the invoice with KSeF e-invoicing comes out of the same system.

Commitment from 6 months
Trial in hoursFirst periodCommitment from 6 monthsRenewals
SettingRangeHow it works
Billing period1 to 3650 daysAny period length in days, each with its own discount.
Trialup to 2160 hoursThe client tests the card on their own workload before paying for the first period.
Minimum termfrom 6 monthsCommitment for dedicated servers. You can price longer terms lower.
Setup feeyour amountA one-off fee added to the first order.
Admin hoursevery started hourHelp from your team with the setup goes onto the client invoice.
Custom offerpayment linkFor an unusual order your staff prepares an offer the client pays from a link.
05Hardware

Machines with cards, run like any dedicated server

A GPU server is still a physical machine in your rack. It gets everything Flaem does for dedicated servers, from intake into inventory to the return.

  1. Machine inventory

    Cards recorded in the hardware profile, and the client sees in the store what they buy and in which location.

  2. Burn-in before sale

    A few hours of full load on memory, CPU, disks and network, with a report full of numbers.

  3. Network install

    The OS installs over iPXE right after payment, without a trip to the data centre.

  4. Controller and console

    Redfish or IPMI, SOL and KVM console, the rescue system and PDUs.

  5. Safe return

    When the contract ends the machine goes to service with a "wipe disks" note and never returns to sale on its own.

06Questions

Questions about GPU servers and AI models

What the client gets, what they see in the panel and how you bill it.

With a dedicated server and a virtual machine with a card, yes, the card works only for them. With an AI model as a service the client uses the model and you manage the cards.

The ones you run on your cards. You decide which models are on offer, which cards they run on and what a package with access costs.

Yes, in their panel: load, card memory, temperature and power draw, next to the controller sensors, CPU, memory, disks and network. You see the same for every machine.

No. The model runs on your cards, so requests and answers stay with you.

Yes. You set a trial counted in hours, up to 2160, in the package. The client tests the card on their own workload before paying.

No. GPU servers and AI models run in the same Flaem installation as games, VPS and dedicated servers, with the same store, invoices and team.

We will show GPU sales on your own hardware

Tell us how many machines with cards you have and what you want to sell: whole servers, virtual machines or access to models. We show the store, card telemetry and billing, then work out the licence from the price list.