Network status Portal Inquiry
GPU compute

AI performance that stays in Czechia.

Over 200 GPUs in our own datacenter. A fixed monthly price instead of a bill that looks different every month — and data stored and processed within the GPU service in the Czech Republic.

200+
GPUs in operation
96 GB
max VRAM per card
100 %
data in CZ
24 h
typical deployment
GPU cards in a server
Available cards

Choose by model size

VRAM is what matters — it determines how large a model fits on the card. We size the rest of the system so the GPU never waits for the CPU or the disk.

RTX 4090

in stock
24 GBVRAM
16 vCPU, 64 GB RAM 1 TB NVMe 1 Gb/s connectivity
9 900 Kč/mo.or Kč 26/hour

Inference of smaller models, fine-tuning, image generation.

Order

RTX 5090

in stock
32 GBVRAM
24 vCPU, 128 GB RAM 2 TB NVMe Shared 10 Gb/s connectivity
14 900 Kč/mo.or Kč 39/hour

Production inference you can rely on. The most common choice.

Order

NVIDIA L40S

in stock
48 GBVRAM, datacenter
32 vCPU, 192 GB RAM 2 TB NVMe Shared 10 Gb/s connectivity
29 900 Kč/mo.or Kč 76/hour

Training and inference, built for continuous operation.

Order

RTX PRO 6000 Blackwell

in stock
96 GBVRAM
32 vCPU, 256 GB RAM 4 TB NVMe Shared 10 Gb/s connectivity
39 900 Kč/mo.or Kč 99/hour

Large models that fit nowhere else, without splitting across cards.

Order

Prices exclude VAT per single card. The monthly rate applies with a commitment of one month or more, the hourly rate for short-term rentals. Multi-card systems are quoted individually.

Storage

Datasets belong next to the card, not behind the internet

There is no point in pulling training data over the network every epoch. We attach high-capacity storage directly in our datacenter, connected locally — so the GPU never waits for transfers.

STORAGE PRICING
CapacityTypePrice/mo.
4 TBNVMeKč 4 000
10 TBNVMeKč 10 000
25 TBSATA SSDKč 20 000
50 TBSATA SSDKč 40 000
100 TB and morecustomon request
Use cases

What our GPUs run

Inference of your own model

The model runs continuously and serves your application. The most common deployment — fixed price, no token counting.

Model fine-tuning

Fine-tuning an open model on your data so it speaks your language and knows your industry.

Document processing

Text recognition, data extraction from contracts and invoices, sorting of incoming mail at scale.

Computer vision

Quality control in manufacturing, camera footage analysis, real-time defect detection.

Image and video generation

Visualizations for architects and studios, bulk preparation of product photography.

Scientific computing

Simulations and modelling for research teams that do not want to wait in a queue for a shared cluster.

Private LLM

Your own language model that never leaves our datacenter

For companies that cannot send data to foreign clouds — law firms, healthcare, public administration or anyone under NIS2. We deploy an open model on a dedicated card, connect it to your documents and you get an API and a web interface. Data stays with us, is never sent anywhere and is never used for training.

Request a private LLM
INCLUDED IN DEPLOYMENT
  • Selection and deployment of a suitable open model
  • Connection to your documents and knowledge base
  • API compatible with common tools
  • Web interface for your employees
  • Fixed monthly price based on the chosen card
  • No training on your content
Dedicated clusters

When one card is not enough

For long-term training or production workloads with predictable demand we build dedicated systems — multiple cards in one machine or interconnected nodes with a fast internal network. Capacity is reserved for you alone and is never shared.

Request a cluster
BUILT TO ORDER
  • 2 to 8 cards in one machine
  • Interconnected nodes for larger training runs
  • Shared storage for the whole cluster
  • Connection to your infrastructure over a private network
  • Contractually guaranteed capacity availability
Why us

When local GPU infrastructure makes sense

Eron X IT

  • Data physically in the Czech Republic
  • Fixed monthly price you know in advance
  • No egress fees for data sent out
  • Latency of a few milliseconds from all of Czechia
  • Support in Czech, with a technician, not a form
  • Storage local to the card
  • We are in NIX and Peering, with direct peering to CESNET and Cloudflare — data in Czechia, Slovakia, Austria and Germany moves extremely fast

Typical foreign cloud

  • Data outside the EU or in an unspecified location
  • Bill changes with usage
  • Fees for downloading your own data
  • Latency across half a continent
  • Support via an English-only ticketing system
  • Storage as a separate paid service

Not sure which card fits your model?

Write to us about what you want to run — we will recommend a system and prepare an offer. The consultation is free.

Send inquiry
FAQ

GPU compute in practice

How quickly can I get a card?

Cards in stock are usually deployed within 24 hours. For multi-card systems and clusters we confirm the timeline when you send the request.

How do I connect to the card?

You get access over SSH and optionally a web interface. We prepare the environment according to what you work with.

Can I later upgrade to a stronger card?

Yes. Switching to another card is handled without penalties — you only pay the difference from the next period.

Is the card exclusively mine or shared?

Dedicated. You do not share it with another customer and you never compete for performance.

What if I only need a card for a few days?

We also offer an hourly rate. For longer projects, however, the monthly price pays off significantly more.

Where exactly is my data?

In our datacenter in the Czech Republic. We do not resell third-party capacity, so we know exactly which machine you run on.