Both the application of high-performance computing in the automotive industry and the challenges of trying to get better performance out of HPC systems will feature at this year’s ISC High Performance Computing conference to be held in Frankfurt, Germany, in July. Both location and date have changed this year – previous conferences have been held most recently in Leipzig and have taken place a month earlier in June.
The opening keynote will review the application of high-performance computing to the automotive industry and, in particular, how Mercedes-Benz makes use of the technology. The talk will be given by Jürgen Kohler, head of NVH CAE and Vehicle Concepts at Daimler AG, who is also chairman of the Prace Industrial Advisory Committee.
In the article Geography helps HPC go green, published on the Scientific Computing World website last month, it ws reported that BMW did all the simulation and computational work for its i-series of energy efficient cars using HPC servers located in Iceland, motivated by a desire for the computation to be energy efficient (and by the low cost of power in Iceland). However, Dr Kohler is expected to discuss Mercedes-Benz’s use of locally placed high-capacity computer clusters in its car development work.
He will point out that digital prototyping has been integral to the work of Mercedes-Benz car development for more than a decade. Functional requirements such as active and passive safety as well as driving, acoustic and thermal comfort, frequently entail trade-offs not only among each other, but also with other aspects such as weight, cost and manufacturability. Consequently, in the quest for the optimum ‘brand-appropriate’ compromise, the digital prototype has proven to be a major boon.
But the computing models are constantly getting bigger because they now also map localized details geometrically and physically. To ensure that this does not impact the turn-around times in day-to-day development work, rapid numerical algorithms and cost-optimised high-performance computers are deployed, that can keep pace with this trend and are designed for high-capacity computing. It is for this reason, that Mercedes-Benz is using locally placed high-capacity computer clusters. He believes that efficient and failsafe operation will ensure a reliable vehicle development process.
On Tuesday, 14 July, one of China’s most eminent HPC specialists, Professor Yutong Lu, will explore the need to make applications run more efficiently on high-performance machines. Professor Lu is the director of the system software laboratory at the school of computer science at the National University of Defense Technology, China, and led the design of the Tianhe-1A and Tianhe-2 systems, the latter being currently the most powerful computer in the world.
However, despite the improvements in hardware, Professor Lu will point out that the efficiency of real applications is still limited by conventional models, algorithms, and IO processing. The scalability and efficiency of application software need to be improved to take advantage of the capabilities of emerging supercomputers. She will discuss the capabilities of CPUs, accelerators, interconnect and I/O storage systems and how they fit together to create a whole HPC-system. She will also discuss large-scale applications running on Tianhe-2, including CFD, genomics, and cosmology, and how they are designed and supported by balancing the system’s computation, communication, data management, and fault tolerance. In her view, a co-design approach is needed for research and development activities, to deliver a whole system for scalable computing, and to support the large-scale domain applications efficiently.
The following day, Professor Thomas Sterling will deliver an in-depth presentation of progress in the critical area of runtime system software, which will play an increasingly significant role in supercomputing of complex and highly-scaled heterogeneous computers. In his presentation, Sterling will examine application drivers stressing the boundaries of system performance and memory capacity. Sterling is professor of Informatics and computing, in the school of informatics and computing at Indiana University and chief scientist and executive associate director of the Center for Research in Extreme Scale Technologies (CREST).
