Trek Bikes has turned to high performance computing to help develop new bicycles, and the effort is seeing results.
Early on, convincing Trek officials to try HPC wasn't easy sell for Mio Suzuki, an analysis engineer who's been at the bicycle maker since earlier this decade, starting as an intern. Suzuki brought to Trek a knowledge of HPC - and a lot of motivation. "Trek needs to be doing what nobody else has done in the industry," she said.
Trek had been using 10 "OK-grade CPUs" on a server, which Suzuki says provide insufficient power for the computational fluid dynamics (CFD) application used by cycle developers.
Suzuki, who holds a master's degree in engineering physics from the University of Wisconsin-Madison and has experience using cloud-based HPC systems, faced a challenge in her quest to convince Trek colleagues about the merits of cloud-based CFD analysis, and its ability to reduce the need for physical wind tunnel testing.
"I'm kind of my own IT person because there is no other person in Trek who has HPC experience and knowledge," said Suzuki.
About a year and half ago, Suzuki began running simulations on a cloud-based cluster operated by R Systems, an HPC cloud provider in Illinois. Trek now uses 128 cores. It has reduced the amount of time needed to run computational fluid dynamics (CFD) simulations from 13 hours to about 40 minutes, enabling Trek to run more simulations and thus increase the analysis of its bike designs.
Additional HPC resources has made it possible to reduce the use of a wind tunnel for testing, which consequently cuts costs.
"But initially, because there was no one else who knew anything about HPC at Trek, I faced quite a bit resistance, and questions, internally," she said. Once she showed the 90% reduction in analysis time, resistance to HPC gave over to "why are we not doing more?" said Suzuki.
The major benefit of HPC at Trek is "a lot more testing virtually" to build a better bike. "It's really accelerated the pace of analysis at Trek," she said.
At the SC13 supercomputing conference last week in Denver, Trek was held up as an example of a mid-sized manufacturing firm successfully using HPC. Proponents of the technology contend that there are tens of thousands of manufacturing firms that could benefit from HPC, but have yet to do so.
Today, HPC is largely seen as the provenance of large firms and big money.
But the advent of cloud-based HPC systems, as well as the increasing capabilities of in-house systems, is expanding the technology's use in product development to a broader range of firms.
HPC analysis, such as CFD, may not completely replace costly wind tunnel testing, but it can change a wind tunnel's role from fashioning a design to confirming one made virtually. HPC can also reduce the number of physical prototypes needed to make a finished product, which can speed time to market.
Raymond Bair, chief computation scientist at Argonne National Laboratory, who helped assemble the lineup of conference presentations, says "there hasn't been much focus on the real, practical end use of HPC," he said. By telling the stories of firms such as Trek, the hope is to "inspire other companies" to investigate HPC.
Walter Kirchner, chief technologist at the Council on Competitiveness and an engineer at Argonne on loan to the non-profit group, says many firms outside of the high-tech and aerospace industries "aren't near up to speed in terms of what they can with computing to enable them to innovate to get to market faster, and to virtually prototype things."
Kirchner said HPC is underpinning gains in jet engine fuel efficiency, and is also playing a major role in energy exploration through seismic imagining.
However, Kirchner believes Congress doesn't realise HPC's role, its potential for the economy and how investments can pay off.
"If we do exascale, we will make affordable, abundant petascale," said Kirchner, adding that lower cost petascale systems "enable a step change in manufacturing and commerce."