NVIDIA DGX,
supplied in Europe
DGX Spark and DGX-class systems on one European contract: hardware, licences and, if you want it, the team that deploys it. Invoiced from Vienna, no import for you to clear.
Everything below is available to order. We do not publish prices because the configuration, the licences and the term change the number; you get a written quote instead.
| System | Memory | Form factor | What it is for |
|---|---|---|---|
| DGX Spark | 128 GB unified, 273 GB/s | Desktop | A local machine for models that need capacity more than speed. Up to three pair by cable, four through a switch. |
| DGX-class servers | HBM per card, 8-GPU configurations | Rack | Training and heavy multi-user inference, where NVLink between cards is the point. |
| H200 NVL · H100 NVL | 141 GB · 94 GB HBM | PCIe card | Adding DGX-class memory to a server you already have. |
| NVIDIA AI Enterprise | – | Licence | The supported software layer. Included with some systems, separate with others. |
All of the above is available to order. Lead time is confirmed with every quote.
Hardware alone is a fine order
You do not need a project, a licence bundle or an engineering contract to buy from us. One DGX Spark on a desk or a row of DGX-class servers in a rack, any quantity: we quote exactly what you ask for, at a competitive price, invoiced in the EU. The solutions are here when you want them, not a condition of sale.
One counterparty in Vienna. No cross-border purchase to arrange, no import to clear on your side.
Where a card needs a licence to do the job you bought it for, it is in the quote, not discovered three months later.
Our partner Vixen.UNO designs, deploys and supports it. Under the same contract, or not at all – your call.
Manufacturer warranty handled through us, with support alongside the team who runs the system.
Most DGX Spark buyers want the memory for 70B-class models. Worth knowing first: a dense 70B in FP8 measured 2.7 tokens per second on one user, because 70 GB of weights has to be read through 273 GB/s for every single token. The capacity is real. The throughput has a ceiling.
Read the full analysis →A DGX H100 draws up to 10.2 kW. Four of them in one rack is about 41 kW, more than three legacy racks combined. A 400 V 32 A three-phase feed gives you roughly 22 kW, and you plan at 80% of that.
Read the full analysis →Weights are one multiplication. The KV cache is the half that surprises people: a 70B in FP16 at 128K context is 140 GB of weights plus 40 GB of cache, for one user.
Read the full analysis →Do you sell DGX Spark and DGX servers, or only one of them?
Why are there no prices on this page?
Is NVIDIA AI Enterprise included?
Can you deploy it, or only ship it?
Where does it ship from and who invoices?
Measured token rates, the 273 GB/s ceiling, and the gap between what the slides promise and what owners see.
Read the article →Rack density numbers, the airflow arithmetic and the four checks that decide whether your delivery can be switched on.
Read the article →Weights, KV cache and the overhead you only see in logs: the working formulas.
Read the article →We reply within one business day