Bare-metal nodes
Full nodes with up to 8 B300 GPUs over NVLink, CPUs with up to hundreds of cores and terabytes of RAM: the entire machine, no neighbors.
EMT Compute · by EMT Aerospace Technologies
Bare-metal nodes, single GPUs and APIs for training, inference and scientific simulation. The infrastructure behind EMT’s programs, open to researchers, companies and laboratories.
EMT headquarters · LumezzaneActive GPU nodesIllustrative orbits
Last updated —Telemetry temporarily unavailable: full data is available at emtcompute.com.
Available GPUs by model
Service status
Peak performance is a dense BF16 estimate calculated from manufacturer specifications for the GPUs currently available.
Full nodes with up to 8 B300 GPUs over NVLink, CPUs with up to hundreds of cores and terabytes of RAM: the entire machine, no neighbors.
From the RTX 5090 to the H200: launch a GPU in minutes, with PyTorch or CUDA templates and SSH access.
Homogeneous node groups on InfiniBand or RoCE networks for large-scale distributed training.
Describe what you want to build: the system estimates data, memory, time and cost, and selects the most suitable resource.
Ready-to-run codes for fluid dynamics, molecular dynamics, quantum chemistry, high-energy physics, structural analysis and robotics.
Catalog, quotes, orders and metrics via API, with scoped keys, expiry and spending caps, and idempotent orders.
Designing a rocket engine, training an artificial intelligence and making a humanoid robot move all require the same resource: massive computing power, available when needed. EMT Compute was born from the needs of our own programs.
# catalogo pubblico, in tempo reale
$ curl -s https://emtcompute.com/api/v1/catalog \
| jq '[.nodes[] | {model, count, city}] | .[:3]'
[
{ "model": "B300 SXM6 AC", "count": 8, "city": "Helsinki" },
{ "model": "H200", "count": 8, "city": "Kansas City" },
{ "model": "H100 80GB HBM3", "count": 8, "city": "Dallas" }
]
$ ▌
EMT Compute is the GPU cloud of EMT Aerospace Technologies, available at emtcompute.com: single GPUs, full bare-metal nodes and APIs for training, inference and scientific simulation, billed for actual usage time only.
The catalog includes GPUs from the RTX 3090, 4090 and 5090 to the A100, H100, H200, B200 and B300, in data centers worldwide. It shows only machines available at that moment and updates automatically.
You top up a credit balance in US dollars, paying in your own currency. At launch, the amount for the maximum selected duration is reserved, but you pay only for actual usage time; credit does not expire. Rates and taxes are listed at emtcompute.com.
No. Guided training estimates data, memory, time and cost and recommends the most suitable resource; experts have access to every parameter and recipe for distributed training.
From almost anywhere in the world. Countries and parties subject to international sanctions are excluded.
Sign up at emtcompute.com and launch your first GPU in minutes.