Empiria Insight

Hardware for your own AI

Your models.
Your computer. Your operation.

I build and configure AI computers for your company — for local inference, experiments and fine-tuning. Choose the specification; I check whether it fits technically and prepare your quote.

Configure a computer Performance reference values ↗
32 CPU coresG4 or G5Up to 4 GPUsAssembly on request

Configure a computer and request a quote

Your specification

Configure.
Receive a quote.

01 Platform
02 Memory

G4: 32 or 128 GB DDR4. G5: 32 or 192 GB DDR5.

03 Additional storage

RAID capacities are the sum of the drives. Usable capacity and RAID level are set in the quote.

04 GPUs

RTX6000 means the NVIDIA RTX PRO 6000 Blackwell with 96 GB. Whether the Workstation or the Max-Q edition is fitted, and how the software fits your project, is checked in the quote.

Your project

What do you want
to use the computer for?

The specification you chose and the price date shown are sent together with these details. The request is without obligation and does not place an order.

Contact details and requirements

* Required. Please do not enter confidential data or credentials. How this request is processed is described in the privacy notice (German).

From a choice to your own system

01

Review the project

We clarify models, memory needs, software and the operating environment. Your configuration is the starting point.

02

Agree the quote

You receive a concrete quote with components, assembly, setup and lead time. Only then do you decide whether to commission it.

03

Build and hand over

After you commission it, I build and test the agreed system. Setup, integration and handover are arranged to suit your team.

Think about hardware and data
together.

If the use case or the data basis is still open, I can help you record and assess what you have, and with a pilot and fine-tuning.

Data consulting ↗

About the configuration

Availability, exact parts, cooling, power supply and suitability for your models are checked before a quote. The RAID capacities stated are gross figures; a particular model speed or noise level is not promised here. Reference values from published benchmarks place the specifications by speed and cost per million tokens.