Konni is a model of the Berlin Supercomputer Cray XC40 'Konrad' build from 10 Raspberry Pi 3 microcomputers to demonstrate massivly parallel computing.

In 2017 Konni got a complete redesign building a 1:25 scale model. After measurement process, CAD and 3D printing in dozens of versions Konni now is a real clone of Konrad. Konni even shares the transversal cooling concept and power rail with its bigger brother.

HLRN Konni 2.0
Konni 2.0

Konni in front of Konrad Raspberry Pis inside of Konni
Konni in front of Konrad and Raspberry Pis inside of Konni

The Hardware

The decision to use Raspberry Pis is based on the idea of showing how simple a small parallel computer can be built with a small budget. Konni is based on its sibling TinyTitan from Oak Ridge National Laboratory.



After the redesign in 2017 Konni also got an own built power distribution and power control unit with air-cooling, besides WLAN for remote access.

If legit, Konni could reach a rank in the lower half of the Green500 ranking of the most energy-efficient Supercomputers, if it would be one.

A detailed partlist is available in KonniPartList.

Videos of the 3D Print

3D Print timelapse of one of the cabinets of Konni and the doors with inlayed hinges and a colored CRAY logo

The Software

Konni uses TinySPH from Oak Ridge National Laboratory, a parallel 2D Smoothed Particle Hydrodynamics (SPH) code. SPH is used in many fields of research including Astrophysics, Ballistics, Volcanology and Oceanography. The application demonstrates real-life distributed computing. Using Message Passing Interface (MPI) - like on a real supercomputer - the problem is split into 40 subproblems, each handled by one physical core of Konni. Each core is represented by its own color.

Besides worker nodes we've integrated a dedicated head node with monitoring capabilities. Thus a WLAN device like a tablet is capable of monitoring the load of the cluster using Grafana.


Created by ChristianSchimmel & AlexanderKammeyer & SaschaFrick & TobiasWatermann

  • 20180112_121530.jpg:

Norddeutscher Verbund für Hoch- und Höchstleistungsrechnen
Back to top of page