Event Summary
Compaq Computer Corporation and the Pittsburgh Supercomputing Center (PSC) in Pittsburgh, PA announced that the National Science Foundation (NSF) in Arlington, VA has selected them to build and manage the world's largest supercomputer to offer scientists access to a wide range of non-military, scientific applications. The contract, valued at an initial $36 million for hardware, software and services, will be managed for NSF by PSC, with first delivery of systems expected by November 2000.
Based on Compaq's AlphaServer SC architecture, the supercomputer will consist of 682 quad-processor Compaq AlphaServer systems with a total of 2,728 processors that will deliver peak computing power of greater than six trillion floating point operations per second (TeraFLOPS). The system will run Compaq's Tru64 UNIX operating system with 2,728 gigabytes of memory (1 gigabyte per CPU) and 50 terabytes of storage.
Under terms of a joint agreement between Compaq and PSC, Compaq will provide the hardware, software, performance and benchmarking activities while PSC will do the porting, tuning and development of efficient parallel applications.
This supercomputer system will allow NSF to establish a single, new terascale computing system to enable U.S. researchers in all science and engineering disciplines to gain access to leading-edge computing capabilities to look at such things as the structure and dynamics of proteins useful in drug design, storm-scale weather forecasting, and the modeling of earthquakes and global climate change.
Background note: The Alpha chip and architecture was designed by Digital Equipment Corporation, which was acquired by Compaq in 1998.
Market Impact
Winning a high-profile contract such as the Pittsburgh Supercomputing Center is a nice trophy, and demonstrates the Alpha chip's power for high-end applications. However, contracts for 2700-processor machines are not going to get awarded every day, so Compaq must concentrate on its regular channels to get the revenues needed.
Compaq announced its AlphaServer GS Series (code-named "Wildfire") in May (It Takes More Than a Fast CPU to Rule the Web), with the stated goal of reaching $1 Billion in sales for the year. Another goal was to make big inroads into Internet infrastructure.
Since then, its sales have been, uh, less than stunning. Current reports indicate that after the initial backlog of 200 systems, only 50 (fifty) additional systems were sold in the quarter. Compared with Sun's E10000 (255 in their first shipping quarter) and IBM's RS/6000 S80 (720 in their quarter), these are hardly impressive figures. (Yes, we realize that backlog + new orders = 250. We discount some of the backlog because Wildfire was over a year late, resulting in more pent-up demand.)
Unless these systems sell for an average of $2.5 Million (we think this unlikely), the sales force will need to redouble their efforts in order to reach $1 Billion. What mystifies us is the persistent lack of enterprise-focused performance data from Compaq - we commented on this two months ago, and have seen little change. Perhaps Alpha Marketing feels benchmarks are unimportant to customers?
The short term impact on Compaq is the potential for having a product not recoup its development costs. Compaq has repeatedly stated that the Alpha is an important part its long-term strategy, so success of the GS series is important from a face-saving point of view. Since Compaq is not big on charity, i.e., keeping a product which is not "pulling its weight", the risk is that the product line will be cut back significantly at the high end. Although volume (through low-end sales) is key to reducing Alpha costs and thus overall cost, a high-end cutback may precipitate a domino effect. We think this would be too bad, because we appreciate a number of Alpha's strengths, but we understand market reality.
User Recommendations
The Pittsburgh Supercomputing Center win shows that Alpha can compete at the very high end of the market. If you have a supercomputer-class task you really need to have solved, such as advanced computational fluid dynamics (CFD) or weather patterns or trying to determine if Alan Greenspan will raise interest rates, the high-end Alpha systems are deserving of your attention. In the supercomputing space, IBM is also a serious contender, and should be investigated/reviewed by potential customers.
The PSC win also demonstrates the scalability of the Alpha architecture. This will help reassure customers who intend to buy smaller systems, but want to plan for expansion.