Start here
What Blackforge is
GPU compute from a forge in Iceland, with one price per plan and a public record of every hour.
Blackforge rents NVIDIA B200 GPUs to AI teams. The GPUs sit in our own forge, F1, on the Reykjanes peninsula in Iceland. You send a container, say how many GPUs it needs and which plan pays for it, and it runs. Every GPU reports its load, heat and power once a second, to you and to the public status page.
What is in the forge
F1 has two nodes. Each node is an HGX B200: 8 Blackwell GPUs with 180 GB of memory each, joined by NVLink 5, and the two nodes are joined by InfiniBand. A job can use 1 to 8 GPUs on one node, or all 16 across both. The full spec is on Forge F1.
Four plans
| Plan | USD per GPU hour | In one line |
|---|---|---|
| Strike | 3.79 | Billed per second, 60 second minimum. Nothing to sign. |
| Hour bank | 3.09 | Buy 1,000 GPU hours at a time. Spend them on any job within 120 days. |
| Anvil | 2.89 | One node, all 8 GPUs, yours for a month or more. 16,640 USD per node per month. |
| Ember | 1.39 | Runs on idle GPUs. 30 seconds of notice before it gives the GPU back. |
You can mix them. Most teams run experiments on Strike, put steady work on an hour bank, and fill gaps with Ember. Teams that need a node for weeks take an Anvil.
Why a forge in Iceland
A GPU turns nearly all of its power into heat. In Reykjanes the outside air carries that heat away for most of the year, and the power comes from geothermal and hydro plants on a fixed price contract. Cheap cooling and a flat power price are where our prices come from. Iceland is in the European Economic Area, so data stored in F1 stays under GDPR.
What Blackforge does not do
No object store product, no managed databases, no serverless functions. There are GPUs, a scheduler, storage mounted into your job, a heat feed and an API. Bring the rest from wherever it already lives.
Where to start
- Quickstart: run your first job in about five minutes.
- Concepts: the dozen words the rest of the docs use.
- Plans: which plan fits which kind of work.
- API: every endpoint and field.