- APE100 (1993-1995)- 25 GFlops
- APE1000 (1999/2001)-144 GFlops peak
- apeNEXT (2005/2006)- 4000 GFlops
... but time flies!
and a new supercomputer is needed.
The making of a Hybrid GPU Cluster (nameless so far)
- a picture story -
directors and producers
... some actors
Markus Klappenbach (cluster administrator)
and transformers: big boxes, smaller boxes, empty space and a christmas tree with own electric power supply
The directors and producers had not only the duty to find components to transform to a GPU-Cluster and to fight for money but additionally to find electric power under very special conditions. But finally the project could get started in a high security area of the university computer center (HRZ).
The empty compartment gets filled - the empty frames get filled under the watchfull eyes of directors and actors. As it is Christmas time even the little, electrical autonomous christmas tree is watching the making of the GPU-Cluster permanently.
A new actor shows up - hidden behind the still empty racks: Christian Schmidt
The project is realized together with sysGen Gmbh (a supplier for computer engineering) and NVIDIA (a leading producer of GPUs).
The content of a lot of smaller and bigger boxes has to be filled into 15 racks. In the end there will be 400 GPUs with a processing power of 500 Teraflops.
- the front side
- the beautiful rear side
What makes a supercomputer really expensive?
Electricity is a problem you should never underestimate!
The power station under construction. The picture on the left shows really impressive power cables for a really impressive computer.
One particular feature of the new computer is its comparatively low power consumption. It is 50 times smaller than a system with the same computing capability composed of PCs.
Besides the installation of the GPU-Cluster a (little but not unexpensive) problem occurred: The bigger and smaller boxes transformed from full to empty boxes. Only with the help of heavy equipment the problem could be solved in a long, long lasting campaign.
The memory of the cluster
After several weeks of hard work the cluster is now nearly finished.
We protect our baby with all means.
- Fireprotection by Gloria.
- We even put the cluster behind bars
- and extra doors
- we save electrical power wherever we can-
Mathias Wagner is starting the cluster with pure manpower on January 25 under the watchful eyes of Olaf Kaczmarek.
Inauguration of the new GPU-Cluster
January 25, 2012
- Prof. Peter Braun-Munzinger
ExtreMe Matter Institute EMMI, GSI, TU Darmstadt and FIAS
Nucleus-nucleus collisions at the LHC: from a gleam in the eye to quantitative investigations of the Quark-Gluon Plasma
- Prof. Richard Brower
QUDA: Lattice Theory on GPUs
- Axel Köhler
NVIDIA, Solution Architect HPC
GPU Computing: Present and Future
- Prof. P. Braun-Munzinger
- Prof. R. Brower
- A. Köhler