May 30, 2005
SGI And ESI Group Make Precise Automotive Simulations Scalable
Please note that contributed articles, blog entries, and comments posted on MCADcafe.com are the views and opinion of the author and do not necessarily represent the views and opinions of the management and staff of Internet Business Systems and its subsidiary web-sites.
SGI And ESI Group Make Precise Automotive Simulations Scalable
Manufacturers today struggle to reduce delays in product design as they work with more and more refined models. To keep up with these time and quality constraints, they look for solutions that can scale with their growing needs. Silicon Graphics and ESI Group announced they have pushed scalability to new extremes, enabling users to run a crash simulation model eight times larger than today's standard industry practice. The achievement represents a significant competitive advantage for automotive designers confronted with ever growing model sizes and shrinking turnaround time.
Today the typical full car simulation model size is 1.1 million elements and can be run overnight. In recent tests, SGI with ESI Group successfully ran a model of 9.1 million elements at 338.20 gigaflops per second performance on a SGI Altix 3700 Bx2 server with 512GB memory running PAM-CRASH 2004 on 512 Intel Itanium 2 processors.
On the high-density Altix system, SGI and ESI Group were able to deliver the crash case results on model with 9.1 million elements - 8x that of a typical model - within 8 hours 11 minutes, enabling engineers to use the results of the full case computation with increased precision as early as the next morning. This means that an engineer can run a model eight times more complicated within essentially the same time as a standard sized model.
This outstanding result is made possible with a complement of the SGI Altix NUMAflex architecture with huge globally addressable shared-memory and PAM-CRASH software's enhanced scalability. The solution enables engineers to cope with the expected growth of model size due to factors like model refinement, increase model fidelity, and modeling material rupture. Additionally, insurance companies rely on simulations to evaluate risks, including specific car-to-car and car-to-passenger interactions. With such a configuration car manufacturers can now address new applications such as the simulation of a detailed car body with trim, or investigate critical zones for rupture prediction.
"While high-performance computing hardware has moved toward scalable systems, ESI Group has invested in a new architecture for its crash test solution solver," said Vincent Chaillou, President and Chief Operating Officer, Product Operations, ESI Group. "The new generation of our parallel solver, first delivered in 2002, offers unprecedented scalability for full car model computations."
"The Altix server family offers tremendous flexibility in configuring for application-specific needs," said Christian Tanasescu, Director of Engineering, Simulation Applications, SGI. "Altix can be used in single system image configurations up to 512 processors or as a supercluster that equips up to 2,048 processors with globally addressable memory. With an industry-leading communication latency of less than 1 microsecond, SGI's NUMAflex interconnect enables PAM-CRASH to scale to dramatic new heights."
Computationally, vehicle crash simulation is one of the most demanding simulation applications because of the sheer number of discrete parts in a vehicle that interact with each other during a crash - forces, vectors, materials, occupants, and on and on. It's a very tall order to simulate this complex behavior, so a lot of computing horsepower is required and SGI's Altix 3700 seems to be the right machine to fulfill this requirement. The Altix 3700 is part of SGI's newest generation of high-end servers with a price point starting at around $100,000.
Since most of us are PC users, we have learned that for the money, more is usually gained by adding memory (RAM) to a system than by adding a faster processor for boosting computational power, and this is even more true in supercomputing. The SGI Altix exploits this with a hugely scalable memory architecture, offering up to 24 TB (terabytes - that's 24 x 1012 or 24,000 gigabytes (GB)!) of globally addressable memory in a system. Contrast that to clustered architectures where addressable memory is typically limited to "a mere" 32 GB.
Speaking of processors, though, an SGI Altix 3700 Bx2 is comprised of nodes. Each node in an Altix 3700 system can contain 16 to 512 Intel Itanium 2 processors, up to 6 terabytes of global shared memory, and 48 XIO buses; that can provide over 3 gigabytes per second of sustained I/O bandwidth. The system also leverages SGI's proprietary NUMAflex global shared-memory architecture for maximum application performance. The latest configuration of the Altix 3700 Bx2 also doubles available bandwidth between Altix so-called CPU bricks (clusters of processors within a node) with SGI's NUMAlink 4 interconnect technology- at 6.4GB/sec and less than 1 microsecond latency.
With microprocessors capable of over one billion calculations per second-less than one nanosecond per operation-fast memory access time is a critical factor for achieving balanced, sustained performance on technical workloads, such as PAM-CRASH. Data crosses over an SGI NUMAlink switch, round-trip, in as little as 50 nanoseconds, compared to 10,000 nanoseconds or more with most commodity clustering interconnects. All things considered, the Altix is powerful and fast.
ESI Group is a pioneer and provider of digital simulation software for prototyping and manufacturing processes that take into account the physics of materials. ESI Group's products represent a collaborative virtual engineering environment that the company calls the Virtual Try-Out Space (VTOS), that enables continuous improvement for virtual (digital) prototypes. By reducing costs and development lead times, the VTOS progressively reduce the need for physical prototypes.
Steel, composites or aluminum energy-absorbing structures can be modeled with strain-rate effects and possible damage. Specific formulations have been developed for the large deformation of frontal offset barriers and the newly developed rupture model addresses the automotive industry's increasing use of high-strength materials.
The PAM-CRASH 2G solver can be run in sequential, shared memory or distributed memory parallel modes, and switched from one mode to another. The distributed memory parallel version shares the same input data. The parallel version also has acceleration factors that can take advantage of multiple computer processors (such as found in the SGI Altix). Based on scalable contact structure and dynamic memory allocation, solver architecture offers performance of up to 50% memory saving.
Outwardly, this looks like a good match - PAM-CRASH 2G requires a lot of computing horsepower and the SGI Altix 3700 can provide it. Look for other computationally-intensive applications to migrate to scaleable and increasingly affordable high-performance platforms.
The Week's Top 5
At MCADCafé we track many things, including the stories that have attracted the most interest from our subscribers. Below are the five news items that were the most viewed during last week.
regular transit use and evaluation.
Dassault Systemes and IBM announced the release of Version 5 Release 15 (V5R15) of their product lifecycle management (PLM) portfolio, comprised of CATIA for collaborative product development, and ENOVIA and SMARTEAM for product data and life cycle management, collaboration, and decision support. Concurrently, Dassault Systemes announced V5R15 of DELMIA for digital development of factory and production processes.
You can find the full MCADCafe event calendar here.
To read more news, click here.
-- Jeff Rowe, MCADCafe.com Contributing Editor.
Be the first to review this article