Configure a System Configure

Compute engineered
around the workload.

We design, build and validate GPU systems for AI development, professional workloads and high-performance computing, from a single workstation on a desk to multi-GPU systems and rack-level infrastructure.

Configure a System
Juros Epoch Dual multi-GPU machine learning workstation
Epoch Dual Up to 2× RTX PRO 6000 Blackwell

Machine learning workstation

Epoch Dual

Juros Epoch Dual multi-GPU machine learning workstation

Epoch Dual

A multi-GPU workstation for local AI development: fine-tuning, running several models at once, and machine-learning work that does not fit on a single card. Built and tested to run under sustained load rather than in bursts.

The full range: machine learning workstations, multi-GPU systems, servers and rack, custom builds and the Core series.

View all systems →

What are you running?

Pick the workload. GPU memory, host CPU, memory bandwidth, storage and thermal headroom all follow from it.

Suggested

Epoch One or Epoch Dual, sized by the memory footprint of the models you serve.

Configure

More than assembled parts.

Every system is hand-assembled, stress-tested and optimised for the workload it will run before it ships.

Workload

What you run sets everything that follows.

Configuration

Specified to that workload, not to a stock list.

Validation

The configuration is checked as a whole before parts are ordered.

Assembly

Hand-assembled, not picked from stock.

Stress test

Run under sustained load on the bench, not in bursts.

Quality check

Inspected against the build sheet it was specified from.

Shipped

Packed and dispatched to you.

Designed for sustained load

Conceptual system architecture

CONCEPTUAL, NOT TO SCALE WORKLOAD CPU + SYSTEM MEMORY PCIe / PLATFORM GPU 0 GPU 1 GPU 2 GPU 3 GPU COUNT PER CONFIGURATION AI / ML COMPUTE POWER STORAGE / I/O NETWORKING AIRFLOW → → EXHAUST
Conceptual architecture. No lane counts, capacities, wattages or airflow figures are shown. None are documented yet.

GPU topology

GPU count, spacing and platform layout designed around the workload.

Memory architecture

System memory and GPU VRAM considered together against model and dataset requirements.

Power delivery Conceptual

Power capacity and distribution engineered around sustained CPU and GPU load.

Thermal design Conceptual

Airflow and component placement designed to hold performance through extended workloads.

Storage & I/O

Storage and data-movement architecture selected according to workload needs.

Items marked conceptual describe design intent. Juros' documented process covers configuration to specification, hand assembly, stress testing and workload validation; power and thermal procedure are not yet documented.

What we build with

Component platforms in current Juros systems

NVIDIA

  • RTX PRO 6000 Blackwell
  • H200 NVL
  • RTX 6000 ADA
  • GeForce RTX 5090 / 5080 / 5070

AMD

  • Threadripper PRO 9000WX
  • Threadripper PRO 7000WX
  • Ryzen 9000 series
  • Ryzen 7000 series

Memory & storage

  • DDR5
  • DDR5 ECC
  • M.2 NVMe

Juros builds with these components. No affiliation, partnership, certification, authorised-reseller status or endorsement is implied.

Build around your workload.

Tell us what you need to run. We will size the system around it.

Configure a System