AHPCRC Heads West: Who Wins, Who Loses?

After 17 years in partnership with the University of Minnesota (and others), the Army High Performance Computing Research Center has relocated to Stanford.
Tabor Research analyzes the decisions behind the move as we look at AHPCRC’s westward journey.

A university-industry-government consortium led by Stanford University recently won a $105 million, five-year contract with the U.S. Army to assume
the management of the Army High Performance Computing Research Center (AHPCRC). The AHPCRC is an on-going collaboration to improve Army military
technology and basic research through the application of High Performance Computing (HPC). From its inception in 1989 until April of this year,
the AHPCRC had been managed by Network Computing Services, Inc. (NCSI) and the University of Minnesota in Minneapolis (UM).

The new Stanford-led consortium includes Morgan State University, New Mexico State University at Las Cruces, the University of Texas at El Paso, and
NASA. High Performance Technologies Inc. (HPTi), a private firm based in Reston, Virginia, will provide administrative, technical, and infrastructure
support, including acquiring, installing, operating, and managing HPC systems for the AHPCRC. Charbel Farhat, a professor of mechanical engineering
at Stanford and member of the Stanford School of Engineering’s Institute for Computational and Mathematical Engineering, will direct the program,
which will be housed at the NASA Ames Research Center in Mountain View, California, ten miles from the Stanford campus.

The initial grant will be for five years, with the potential to renew for an additional five years and $105 million. The grant includes funding for
basic research; technology transfer, educational programs, outreach, and support of the HPC systems; and the acquisition of new HPC resources.
According to the Army’s requirements, the $15 million annual allocations include $2 million for research, $1 million to operate and support existing
HPC resources, and $1.5 million for education and outreach. New resources may be acquired with another $30 million over the first five years.

In its research, the AHPCRC emphasizes simulation over experiment and theory, with the aim of developing new technology for the U.S. Army’s use. The
program focuses on four areas of research: to develop new, lightweight materials for military vehicles and equipment; to improve wireless battlefield
communication; to speed the detection of biological or chemical attacks through computational research in nanotechnology and bioscience; and to
stimulate innovations in supercomputing. There is also a strong emphasis on transferring newly developed technologies to non-military applications,
such as lightweight materials used in vehicle manufacture.

UM had been the prime contractor of the consortium managing AHPCRC since its inception in1989. This consortium included Clark Atlanta University, Florida
A&M University, Howard University, Jackson State University, and University of North Dakota. NCSI had been the administrative, technical, and
infrastructure support contractor.

After 17 years of funding, the UM facility has already been liquidated, and all of the major equipment has been shipped to California. This includes
a Cray XT3 (configured with 1158 AMD Opteron processors), a liquid-cooled Cray X1E (1024 vector processors), and a Liquid Computing LiquidIQ system
(40 AMD Opteron processors). A smaller air-cooled Cray X1 (64 vector processors) and Atipa Systems cluster (150 AMD Opteron processors) were decommissioned.
Some older Army property including a StorageTek robotic tape device and a powered-off Thinking Machines CM-5 remain in Minnesota.

Tabor Research Analysis

Moving its resources from Minnesota to California is a major shift for AHPCRC, not lightly undertaken. Tabor Research found there to be both political
and technological motivations behind the move.

Architecture: UM has been a continuing supporter of traditional supercomputing technologies, including ongoing use of large-scale vector-processing
machines. The Stanford consortium promised a greater focus on commodity components and clustering. As with many users, AHPCRC may now be convinced
it can adopt computers with distributed memory architectures without a significant sacrifice in capability, and it may view the use of commodity
components as a more forward-looking strategy.

Partnerships: Two sources close to the decision indicated that the structure of the partnerships within the consortia was a major consideration. The
Army preferred the relationship between Stanford and HPTi – HPTi supports a number of contracts, is on board as a full partner, and is contributing
its own resources – and the organization structure is bolstered by the close connection to NASA Ames. In contrast NCSI appears to have had the
Army center as its only client.

Political influence: Minnesota Representative Martin Sabo, a long-time champion for U.S. supercomputing and UM’s work with the AHPCRC, exited his office
in January after announcing that he would not run for reelection in 2006. Although local business influences were never cited in the decision to
move to California, it is noteworthy that UM is near Cray’s facilities in Eagan, Minnesota, whereas Silicon Valley is home to Intel, HP, Sun, and
a host of other companies, both large and small, that have fueled the cluster revolution in HPC.

To the Victor Belong the Spoils

AHPCRC views the Stanford proposal as providing a unique combination of industry, academia, and government, particularly with the involvement of NASA.
More importantly, the Army predicts the research will be innovative, flexible, and able to meet ever-evolving needs. The educational program at
Stanford has been engaged throughout the process, and AHPCRC is pleased with the strengths of the consortium’s partnerships.

Stanford already has a history as a spawning ground for high-end technological innovation. Tabor Research believes that the acquisition of AHPCRC resources
will establish Stanford as a center of HPC excellence on an ongoing basis. AHPCRC will contribute to and benefit from the Silicon Valley ecosystem
of technology and talent.

The unfortunate losers are those who worked at UM or its partner universities. The dwindling fortunes of Cray and SGI over the last decade have reduced
the number of HPC opportunities in Minnesota, and talented engineers may have to relocate or change industries.

Seen over the longer span of government contract periods, the movement of supercomputing site support between institutions and regions is neither uncommon
nor unexpected. No organization can claim a lock on government research funding indefinitely. Contracts are periodically rebid to allow competing
organizations to present fresh ideas and to allow the government to stimulate efforts in different regions. In this case, AHPCRC has redefined
the three most important words in real estate: innovation, innovation, innovation.

Posted in