The Cray-T3E range of computers were manufactured from 1996 until 2002. The T3E continued the use of large numbers of commodity processors and distributed memory. Evolving from the T3D the T3E was a stand-alone computer which ran both parallel and single node applications concurrently. Around 100 of these machines were manufactured, and many customers migrated their workload from the Cray-T3D to Cray-T3E. The Cray-T3E was available with multiple processor configurations from 64 to 1024 processors. Over the life of the Cray-T3E the processors were available as 450, 600, 900 and 1200 MHz DEC alpha processors. Smaller systems were directly air-cooled, and larger systems used an external liquid heat exchanger.
A 1480-processor T3E-1200 was the first supercomputer to achieve a performance of more than 1 teraflops running a computational science application, in 1998.
The T3E was a scalable system. As more processing elements are added more cabinets are arranged together. Smaller configurations could have direct air cooling.
The original T3E (retrospectively known as the as the T3E-600) had a 300 MHz processor clock. Later variants, using the faster 21164A (EV56) processor, comprised the T3E-900 (450 MHz), T3E-1200 (600 MHz), T3E-1200E (with improved memory and interconnect performance) and T3E-1350 (675 MHz). The T3E was available in both air-cooled (AC) and liquid-cooled (LC) configurations. AC systems were available with 16 to 128 user PEs, LC systems with 64 to 2048 user PEs.
Cray T3E, Dedicated to Exploring Enigmas of Nature, Comes to Lab
|By Jeffery Kahn, firstname.lastname@example.org|
September 19, 1996
BERKELEY, CA — Continuing its mission to extend the limits of scientific understanding, Ernest Orlando Lawrence Berkeley National Laboratory (Berkeley Lab) has acquired a Cray T3E, one of the most powerful supercomputers in the world.
The Cray T3E becomes the centerpiece of the Lab’s National Energy Research Scientific Computing (NERSC) center, a facility that provides high performance computing to thousands of Department of Energy researchers all over the world. The range of energy research problems they work on include fusion, the modeling and design of biological molecules, the development of new materials, and global climate change.
At the leading edge of the revolution taking place in supercomputing hardware, the T3E is a “highly parallel” system capable of teaming thousands of microprocessors. These microprocessors, made by the Digital Equipment Corp., are the world’s fastest. Each DECchip 21164 processor can perform 600 million calculations or MFLOPS per second. Berkeley Lab’s T3E will begin service with 128 processors with plans to scale-up to 512 processors early next year. At that time, the machine will be capable of more than 300 billion calculations per second. A person using a hand-held calculator would take some 40,000 years to do what this Cray does in one second.
Bill Kramer, head of NERSC’s High Performance Computing Department, said this is the first of the new Cray T3Es “that has passed a performance test suite consisting of a wide range of scientific applications. And, it is the first T3E planned for real production use right from the beginning.”
The Cray T3E can be considered a form of time machine. Like eyes that can look into the future, the machine makes it possible to pose questions and resolve problems years and years before it would be feasible by any other approach. How can the internal combustion engine be made both more efficient and less polluting? What changes will occur to the Earth’s climate? What shape molecule should be engineered to perform a targeted biological function? Large scale computation centers attack questions like these, at the frontiers of science. And, as they do so, they pioneer the future of computing itself — hardware, software, and networks destined to be increasingly important to our daily lives.
Bill McCurdy, Berkeley Lab’s associate laboratory director for Computing Sciences, is among those who believe that supercomputers have changed the very nature of science. The traditional interplay between theory and experiment now has been joined by a new mode of inquiry, that of computational experiment.
“Over the past quarter-century,” observes McCurdy, “a fundamental change has occurred in the way scientists and engineers view computation as a tool of research. In the 1960s, computation was a specialized tool whose application was largely limited to a few disciplines of physics, engineering, and chemistry, and which was widely considered to be merely an adjunct of theory. After a quarter-century of spectacular advances in computing hardware and numerical algorithms, we now commonly speak of experiment, theory, and computation as the three principal elements of modern scientific research. The change in our thinking is dramatically highlighted by discussions of large-scale computational experiments appearing in the scientific literature, side-by-side with the results of physical experiments.”
In many cases, computer simulation or modeling is the only approach available to researchers. Physical experiments may not be possible because they are prohibitively large or small, unfold too quickly or too slowly, or because they cost too much. Researchers, however, can create computational models of physical phenomena and validate them with experiments. Through these models, supercomputers are able to simulate and explore what otherwise can be off-limits.
Researchers all over the nation have lined up to use the T3E. Kramer notes that half of its time has been dedicated to “Grand Challenges,” scientific problems that the federal government designates as national priorities. More than 100 research groups have submitted proposals for use of the remaining time available on the T3E. Despite the enormous computing resources of the machine, the Cray can accommodate only one-tenth of these requests.
Scientists interact with NERSC’s supercomputers through the Department of Energy’s major high-speed network, the Energy Sciences Network or ESnet. ESnet’s operations are headquartered at Berkeley Lab.
Like a virtual office hallway, ESnet connects users all over the world to Berkeley’s supercomputers as well as to a number of other unique Energy Department facilities. Because of its need to move huge streams of information, the ESnet is a prime shaper of the future face of the Internet. ESnet is the first national production network to make use of the new Asynchronous Transfer Mode (ATM) technology, which can transmit both voice and data. Some legs of ESnet’s ATM network run at speeds of up to 155 million bits per second. That’s compared to the 28 thousand bit-per-second speed of modems now coming into use on PCs.
Berkeley Lab conducts unclassified scientific research for the U.S. Department of Energy. It is located in Berkeley, California and is managed by the University of California.