Ed McCracken discusses CRI/SGI merger

SGI/Cray logo

by Sam Milosevich

CRI approached
SGI--to cooperate
rather than compete.

In a general session of the North Carolina CUG, Ed McCracken, who was awarded the U.S. Medal of Technology in 1995 and is chair and chief executive officer of Silicon Graphics, spoke about the new Silicon Graphics, Inc./Cray Research, Inc. (SGI/CRI).

A brief history of SGI

SGI, started in 1980 and incorporated in 1982, originally focused on applications-oriented silicon chips. The Defense Advanced Research Projects Agency (DARPA) "geometry engine" led to SGI's focus in 1983 on being a computer terminal company. In 1984 that focus was shifting to workstations, and in 1985 the company went public. Products in 1986 focused on the desktop. In 1988, SGI introduced the Power Series servers, which became the Challenge line in the 1990s.


Market segments

SGI is aggressive and reactive to market demands; rapid introduction of high technology is the company's hallmark. SGI pursues four markets.

  1. Manufacturing, accounting for more than one-third of SGI's business, in areas such as automotive and aircraft

  2. Defense and intelligence, also roughly one-third of SGI's business, focused on image processing and flight or mission simulation

  3. Science, representing more than one-sixth of SGI's business, including weather and environmental, oil and gas, pharmaceutical, physics, and the university market

  4. Tele-entertainment, currently less than one-sixth of SGI's business, although popular with the press

The roughly $4 billion business is split approximately evenly between high-end and low-end performance systems. The high-end business consists of three sectors.

  1. Graphics supercomputers ($0.5B) with slow growth, where SGI has 70% of the market and the competition is mostly "black box" special-purpose systems

  2. Database systems ($0.5B) in large data warehouses and large World Wide Web content warehouses--to manage "rooms and rooms of data" with databases, some for visualization and analysis and some not for analysis

  3. Classic supercomputers ($1B, two-thirds CRI and one-third SGI) for which SGI-CRI has 65% of the market and is twice the size of the nearest competitor

"Just in time" research

SGI itself doesn't do a lot of research; its "just in time research" approach is one of finding solutions from a pipeline of new ideas which SGI then engineers into a solution. SGI's location was chosen to be within a 10-minute drive from Stanford University and other institutions. The DASH project at Stanford led to the SGI Origin product line. SGI is on the lookout for signs of "nonlinear engineering change" potentials.

The supercomputing market is important; moreso is the technical market. The commercial market is not interested in computer architecture as is the technical market.

Why did SGI buy CRI?

CRI needed to broaden their product line and R&D, so CRI approached SGI--to cooperate rather than compete. SGI liked the CRI technology, especially CRI's software technology, algorithmic expertise, understanding of high-end customers, unique "Cray" name, and customer-service activities. While SGI may have been regarded as "SGI wants to sell" to customers, CRI is regarded as "CRI wants to be a partner" with customers.

To address customer service, Mick Dungworth, former chief of CRI Sales and Service, has been appointed vice president of Worldwide Service for SGI and SGI-CRI products. In surveys, the number-one rank for customer satisfaction and loyalty went to SGI in the workstation category and to CRI in the supercomputer category.

The new SGI-CRI wants to partner, to understand customers' business needs, to collaborate; it also wants to find a way to communicate the long-term vision of the future (which SGI historically has done in private) in one-on-one sessions with customers.

Rumors abound

Contrary to rumors that SGI will kill CRI products, especially parallel-vector systems, there are plans to add on to the existing CRI product line with a new generation of the T90 and a new J90 follow-on product. As it happens, there was a relatively similar technology plan for the CRI "scaleable node" and the SGI "Project LEGO."

The ASCI Blue Mountain project was granted to SGI-CRI's scalable node technology with a target of 4,000 processors. This will be approached in phases, starting with the recently-announced Origin 2000, and eventually incorporating vector processing.

Supercomputing at SGI-CRI is just beginning. The year 2010 will bring the possibility of 100 million elements on a chip and one-thousand-fold speed-ups; each technology generation from today will make progress toward that. The issue is software, and that is where CRI comes in. The CRAY Origin 2000 models will start at the point beyond 64 processors in the current Origin 2000 line. The National Center for Supercomputing Applications (NCSA) at the University of Illinois will have a system with 128 processors.


Nonlinear change

Looking for discontinuities, those windows of maximum opportunity, requires working together with the best and brightest of customers, applications developers, and technology developers. The market is being driven more and more to a commodity position, but better money can be made in the first 5 to 10 years. So, the business model is:

"Be there first and charge a little more."

Generating more new products faster is necessary to support a 50% margin and 12- to-13% investment in R&D--and to be distinguished from a commodity markets of perhaps 30% margin and 2% investment in R&D.


The role of CUG, the Cray User Group

The potential value added from investment in R&D enables solving today's problems and it also facilitates looking ahead to the year 2010. The challenge for CUG is to be part of that solution and part of that vision of the future. For SGI-CRI, being good partners means being good listeners. The risk with user groups is driving a company in a linear path.

CUG should relate to all high-end or high-performance computing areas of SGI-CRI. This includes graphics supercomputing and database supercomputing as well as classic supercomputing (which now includes the Origin 2000 product line anchoring the entry point for supercomputing).

CUG should work to expand the applications base, to maximize applications potential of SGI-CRI platforms. And, CUG should add working sessions that target the future between today and the year 2010.

Supercomputing at SGI-CRI is just beginning.

CUG should work
to expand the
applications base
of SGI-CRI platforms.

Contents || CUG home

Comments to: lester@ucar.edu