By Steve Fisher Editor In Chief -- Last week Cray Inc. announced its new SX-6 series supercomputer. As part of our continuing coverage of this new system, Supercomputing Online interviewed Eric Pitcher, Cray’s Director of Government and University Marketing. Supercomputing: Congratulations on the release of the SX-6. Please tell the readers all about it including how it differs from it's predecessor the SX-5. PITCHER: The Cray SX-6 is a scalable parallel vector computer. Each Chassis provides up to 64 GFLOPS of peak performance using only 8 processors. It has 64 GB of uniform shared memory in each chassis, and 6.4 GB/s of I/O bandwidth. Multiple chassis may be coupled together using the IXS Internode Crossbar Switch, which has 8 GB/s bidirectional channels, permitting the system to scale up to 8 TFLOPS. The SX-6 is differentiated from SMP systems because it has high bandwidth memory. Each processor can access memory at the rate of 32 GB/s, for a total aggregate bandwidth of 256 GB/s per chassis. Compared to the SX-5, the SX-6 has about 1/12th of the footprint, uses about 1/5th of the power, and has better sustained performance because of other technology improvements. For example, whereas the SX-5 had 32 chips per processor, the SX-6 has only 1. Supercomputing: Can I have one? PITCHER: Absolutely! Mastercard, Visa, Amex, or Discover? Supercomputing: (laughter) They're all maxed. Moving right along...please describe the relationship between Cray and NEC. PITCHER: Cray has an OEM agreement with NEC that gives Cray exclusive rights to distribute and service NEC vector supercomputers in North America, and non-exclusive rights in most other areas of the world. Supercomputing: How does the SX-6 measure up (as far as performance) with the recently announced offerings from Sun Microsystems and IBM...the 15K and Regatta? PITCHER: On real world applications our experience to date is that the SX-6 is the system to beat for capability-class computing, especially for those applications that do not scale well. Moreover, it is generally easier to achieve high performance from fewer, more powerful processors, such as those found in the SX-6. Cray and NEC continue to focus on sustained rather than theoretical peak performance. Supercomputing: Performance-wise, how does the SX-6 measure up with the Cray SV-1, the soon to be shipped MTA-2 and next year's SV-2? PITCHER: Cray offers multiple platforms because 'one size' doesn't fit all the applications and workloads of our customers. While each of our products may be used to run a general applications workload, the unique architectural features of each product map optimally onto a subset of the thousands of applications that run on our systems. In general terms, the Cray SX-6 and Cray SV1 are traditional vector systems with large shared memory. The SX-6 may be scaled into the Teraflop range (trillions of floating point operations per second). The Cray SV2 will provide extreme scalability up to several tens of Teraflops. The Cray MTA-2 is a multithreaded design in which the CPUs may address up to a Terabyte of uniformly shared memory. Supercomputing: Which system do you view as the "future" of Cray? PITCHER: Our largest development program, with partial funding support from the U.S. Government, is the Cray SV2, which promises to be the world's most powerful supercomputer product. Although the SV2 isn't due out until the second half of 2002, we are already seeing strong advance demand for this system and are confident the first full year of production will quickly be "spoken for" by government, industrial and academic customers. So, the SV2 is a strong bet to bring in healthy revenue. In the near term, we expect healthy revenue from the Cray SV1ex system and are also excited about the potential for the Cray SX-6 supercomputer. Finally, the Cray MTA-2 features a truly revolutionary, multithreaded architecture that could enable a quantum leap forward for some important problems and workloads. ---------- Supercomputing Online wishes to thank Eric Pitcher for his time and insights. ----------