H200 NVL, supplied in Europe
Identical compute silicon to the H100 NVL in existing estates. The difference lives in memory, in faster NVLink bridges and in a 600 W power limit instead of 400 W, and whether an upgrade is worth it depends on how memory-bound your workload is.
- EU contract and invoice
- Licences sized with the hardware
- Engineering, if you want it
- Warranty and support
The card, the bridges and the licence
Against the H100 NVL, the upgrade buys memory, not compute. In Baseten’s tests on SXM systems the gain ranged from none to 3.4×, depending on the workload. We would rather you knew which before ordering.
| Card | Memory | Bandwidth | Notes |
|---|---|---|---|
| H200 NVL | 141 GB HBM3e | 4.8 TB/s | Six HBM stacks. Where the model no longer fits in 94 GB, this is the whole argument. |
| H100 NVL (existing estates) | 94 GB HBM3 | 3.9 TB/s | For comparison, not supplied by us: the card existing estates run, still wider than H100 SXM at 3.35 TB/s. According to a reseller’s end-of-life notice, the last time buy for the H100 NVL was 1 November 2025. |
| NVLink bridges | n/a | n/a | Two- or four-way at 900 GB/s per GPU, against two-way at 600 GB/s on the H100 NVL: the biggest generational difference once a model spans several cards. |
| NVIDIA AI Enterprise | n/a | n/a | A five-year subscription is included with each card; NVIDIA counts the five years from the board’s shipment to the OEM plus 90 days, not from activation. |
The H200 NVL and its NVLink bridges are available to order, with quantities and lead time confirmed in writing with the quote.
Cards alone are a fine order
You do not need a project or an engineering contract to buy from us. One H200 NVL to give a server you already own the memory it lacks, or a set of four bridged together: we quote exactly what you ask for, invoiced in the EU, with quantities and lead time confirmed in writing with the quote.
Not just a box on a pallet
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.
Which one is right for you
After TensorRT-LLM optimisations, NVIDIA’s technical blog of March 2024 states that Llama 2 70B on H200 is limited by compute, not by memory bandwidth. If your workload is in that shape, the extra 47 GB buys capacity, not much speed.
Read the full analysis →The loudest figure in H200 materials is 110× on HPC. The footnote reveals it is four GPUs against a pair of Sapphire Rapids 8480 CPUs, a comparison with processors, not with H100. In another chart H100 ran batch 8 and H200 ran batch 32.
Read the full analysis →A 70B in FP16 with a 128K context is 131 GiB of weights plus 40 GiB of cache: about 171 GiB for one person. The counter-intuitive part: a 7B with multi-head attention spends 512 KiB per token where a 70B with grouped-query attention spends 320 KiB.
Read the full analysis →H200 NVL, in short
H200 NVL, or is our H100 NVL enough?
Why is the H200 NVL 141 GB and not a round number?
Do I need the NVLink bridges?
Is the AI Enterprise licence included?
Can you also supply the server?
From our engineering blog
141 vs 94 GB, 4.8 vs 3.9 TB/s, identical compute: when the upgrade pays, and when an existing H100 NVL is enough.
Read the article →The working formulas for weights, KV cache and runtime overhead.
Read the article →What to check before the servers arrive: the four things that stop a delivery.
Read the article →Check price and availability
We reply within one business day