Infrastructure
NVIDIA HGX B200 in a Tier III datacenter in Madrid
Our own hardware, run by our own team, on a platform certified ISO 27001 and ENS. This is exactly what sits under your workload.
- GPU: NVIDIA B200 (Blackwell) with 180 GB HBM3e and 8 TB/s per GPU.
- Fractions: from 1/8 of a GPU (23 GB), hardware-isolated with MIG, up to 8 × B200 nodes over NVLink.
- Location: Madrid, Tier III. Data only in Spain, no subprocessors outside the EU.
- Certifications: platform UNE-EN ISO/IEC 27001:2023 and ENS Medium; datacenter ENS High.
Spec sheet
Every component, with its number.
NVIDIA's public figures for the hardware and our own operating data for the rest.
| Component | Specification |
|---|---|
| GPU | NVIDIA B200 (Blackwell)180 GB HBM3e · 8 TB/s · native FP4 (NVFP4) |
| Interconnect | NVLink 5 · 1.8 TB/s per GPUGroups of 8 B200 for large models |
| Network | InfiniBand NDR · 400 Gb/sRDMA between nodes; internal traffic never reaches the internet |
| Storage | Exascaler parallel file systemIsolated space per customer, with its own quotas and access control |
| Cooling | Direct-to-chip liquid coolingMore density per rack with less energy spent on cooling |
| Management | NVIDIA Base Command ManagerThe cluster grows node by node with no migrations for customers |
| Datacenter | Tier III · MadridENS High, ISO/IEC 27001, ISO 9001 and ISO 50001 |
MIG fractions
How a B200 is split, and what each part costs.
MIG divides the GPU in hardware: each instance has its own memory and compute, with no neighbours slowing it down. These are the profiles we offer.
1/8 of a GPU
1g.23gb · up to 7
€1.50/h€1,095 /month
1/4 of a GPU
1g.45gb · up to 4
€3.00/h€2,190 /month
1/2 of a GPU
3g.90gb · up to 2
€6.00/h€4,380 /month
Full GPU
7g.180gb · up to 1
€8.00/h€5,840 /month
Responsibilities and certifications
Three layers. Each with its owner and its certificate.
Datacenter
Building, power, cooling and physical security, run by the datacenter operator.
GPU Solutions platform
GPUs, network, storage, isolation and support. Certified by EQA.
Your workload
Your models, data and access. On bare metal, the operating system is yours too.
Security architecture
Your perimeter. Your rules.
Certifications (ISO 27001, ENS Media) are the result, not the product. The product is the architecture that makes them possible. Four layers, auditable one by one.
Full technical documentation available under NDA for your risk assessment.
Your private network, your firewall, your VPN.
Each customer gets a dedicated VLAN / VXLAN. Access comes through your own VPN and passes a firewall with rules you define. No public internet exposure by default.
Key-based identity, full audit.
Authenticated access via SSH keys or corporate OIDC. Endpoints protected with mTLS. Every call is logged: who, what, when, from where.
Hardware isolation, not container isolation.
Your workload runs on a dedicated VM, not in a shared container. Your B200 slice is partitioned with MIG: SMs, HBM3e memory, and cache physically separated.
Encrypted at rest, in transit, residency Spain.
Exascaler with AES-256 at rest. TLS 1.3 in transit. Backups also encrypted. Data doesn't cross the Spanish border, not yours, not its metadata, not the logs.
What we never do
Four things that don't happen, by design.
Operations
Who is behind it.
- Our own team
- Operations, support and consulting from Spain. An engineer answers.
- Under 4 h
- Support response on business days, in Spanish and English.
- 72 hours
- From signature to a dedicated environment in production.
- Status page
- Availability published in real time.
FAQ
About the infrastructure.
How much memory does an NVIDIA B200 GPU have?+–
180 GB of HBM3e with 8 TB/s of bandwidth. That is 2.25 times the memory of an 80 GB H100.
Where is the datacenter?+–
In Madrid, in a Tier III datacenter certified ENS High. All the hardware is owned by GPU Solutions.
Can I have a machine just for me?+–
Yes. We offer full 8 × B200 bare-metal nodes, outside the shared cluster, with your own access.
What happens if a GPU fails?+–
With high availability contracted, we restart your environment on another GPU in the cluster within 24 hours (standard) or 4 hours (24/7). Your data lives on Exascaler, so nothing needs to be moved.
Come and see it
Visit the datacenter and run a demo on B200.