Michael Resch is working in Stuttgart at the High Performance Computing Center (HLRS). In Stuttgart, the automobile industry is a prominent business. In fact, in this German city you will find more machinery than in the USA. HLRS has signed a public-private joint venture with Porsche. The vehicles are developed on supercomputers in order to create a low fuel consumption car.
HLRS is now following a Solution Centers strategy for the automotive, chemistry, energy, and life science industries, together with the colleagues in Karlsruhe. Within 10 years, there will be 1 million electronic cars on the streets. In order to achieve this, you need a sound basis for simulation. Daimler and Porsche are currently using simulation performed on powerful supercomputers.
Michael Resch insisted that we will need more performance to solve the issues. How do we get these Petaflop/s?
Stephan Wesner, his colleague from HLRS, explained that HLRS has established a collaboration with PRACE.
It is essential to give priority to sustained performance of applications - specifically in engineering. The quality of the programming environment is a key enabler for the application developer. The system architecture must support highly scalable applications with a low latency interconnect and a low memory footprint.
The system needs to fit all HLRS customer types. The performance needs to be delivered at the lowest energy costs possible.
Petaflop/s are more than bare metal. The best approach to green computing is to have an efficient code, a
close collaboration with the vendor and the realisation of a local centre of excellence. This can guarantee smooth transitions for the user.
In the collaboration with PRACE, Phase 1 of the HERMES supercomputer, developed by Cray, has now started with an external login server including parallel Filesystem Lustre, a HLRS shared NAS home space, another HLRS server, storage and archive, and a remote visualization server.
The configuration has a peak performance of 1 Petaflop/s and consists of 38 racks with 96 nodes each. There are 96 service nodes and 3552 compute nodes. Each compute node will have 2 sockets and there are nodes with 32 GB and 64 GB.
Cray and HLRS have set up a Cray development centre in Stuttgart with on site staff for production and joint research.
The joined development and definition of the details of Step 2 in Phase 1 are as follows. Step 2 will run in parallel to step 1 realizing an integrated system. The goal is to maintain a similar software stack for both installation steps. Architectural changes are to be expected. There will be a significant increase in sustained application performance. Step 2 is scheduled for Autumn 2013. The overall peak performance of the complete Phase 1 will be more than 5 Petaflop/s.
So, what do we do with all these PetaFlop/s, Michael Resch continued. He gave some examples of Petaflop/s use.
The supercomputers at HLRS need 12 hours to simulate combustion in the combustion modelling in power plants. In process engineering the processes will be improved.
How can we be environmentally safe? This is a big issue for supercomputers. There has to be as little pollution as possible.
In medical engineering, supercomputing is used for bone study to develop artificial hips and shoulders. To this aim, an extreme high level of resolution is required. Simulating an entire patient requires high resolution.
Petaflops are also used in production for people working in the factories. Simulation will increasingly be performed in factories. Augmentive reality has already been used by Mercedes for their G-class type of cars.
Petaflop/s are promising more quality of life for people in Germany and worldwide, Michael Resch concluded. Petaflop/s are necessarey to drive science and engineering. HLRS will focus on sustained performance for engineering applications and beyond. Cray is the partner by excellence for HLRS and will support the development platform for the future. Together HLRS and Cray are working on solutions for the future.