Your slice is physical, not virtual
The chip is split into real parts, each with its own compute and its own memory. What the neighbour does never slows you down.
Dedicated GPU
NVIDIA's most powerful AI chip, whole or in slices, reserved exclusively for your project. In Spain, in euros, from 1.10 € an hour.
Activated on a 10-minute call, no commitment.
how it works
You buy nothing and set nothing up: you reserve power on our machines and it's yours for as long as you need it.
A slice to get going, or the whole card for big projects. By the hour or by the month, at a discount.
A 10-minute call, no commitment: we confirm availability, answer your questions and reserve it in your name.
Spain, EUNobody else touches it: performance is always the same, peak hours included. You give it up whenever you like, from the panel.
what makes it different
On many clouds "your" GPU is split between several customers and performance depends on who you land next to. Here the part you reserve is physically yours.
The chip is split into real parts, each with its own compute and its own memory. What the neighbour does never slows you down.
Because you share with nobody, it runs as fast on a Tuesday lunchtime as at 3am on a Sunday. Easy to plan around.
The machine sits in a Spanish data centre and your data never leaves the EU. No currency-conversion surprises.
how much machine you need
All four carry the same current-generation chip, the NVIDIA B200. The difference is how much of it you reserve for yourself.
For building, testing and small models.
Room to spare for everyday working models.
Half a B200, exclusively yours, for large models.
The entire B200, for training and demanding workloads.
Memory is approximate: the exact split is confirmed when it's switched on, as is the availability of each configuration.
Reserving by the month comes out 10% cheaper than the hourly rate and secures the machine in a market where they are scarce.
how much it performs
The B200 is between 2 and 2.5 times more powerful than NVIDIA's previous chip. Whether you reserve the whole card or a slice, the silicon is the same: what you choose is how much of it works only for you.
against the previous generation
And if you want the exact figure for your case, your technical team sends us the real workload and we send back the measurement in writing.
frequently asked
Multi-Instance GPU: it partitions the B200 into hardware-isolated instances (not software multiplexing). Pick a slice (MIG of 23 or 45 GB) or the whole card.
Yes. The card (or the MIG slice) is yours while you hold the reservation. No shared tenant: predictable performance, no noisy neighbours.
On-demand by the hour, or reserved by the month or year to secure capacity, with a discount on the hourly rate.
A full B200 delivers on the order of 2 to 2.5 times an H100. If you reserve a slice, the silicon is the same and what changes is how much of the card works for you; tell us your workload and we will tell you which size fits.
In euros, per active hour or by monthly/yearly reservation. Same balance as Token Factory and Sandbox, a single invoice.
In our datacenter in Spain. No transfers outside the EU.
You leave us your phone number, we call you in the slot you choose and switch it on together. Ten minutes, no commitment.