The Cydra 5 Departmental Supercomputer: design philosophies, decisions and trade-offs

B.R. Rau, D.W.L. Yen, W. Yen, R.A. Towle
[1989] Proceedings of the Twenty-Second Annual Hawaii International Conference on System Sciences. Volume 1: Architecture Track  
1 groups or departments of scientists and engineers.' I t costs about the same as a high-end superminicomputer ($500,000 to $1 million), but it can achieve about one-third t o one-half the performance of a supercomputer costing $10 to $20 million. This results from using highspeed, air-cooled, emitter-coupled logic technology in a product that includes many architectural innovations. The Cydra 5 is a heterogeneous multiprocessor system. The two types of processors are functionally specialized
more » ... nally specialized for the different components of the work load found in a departmental setting. The Cydra 5 numeric processor, based on the company's directed-dataflow architecture,* provides consistently high performance on a broader class of numerical computations than d o processors based on other architectures. It is aided by the highbandwidth main memory system with its stride-insensitive performance. The interactive processors offload all nonnumeric work from the numeric processor, leaving it free t o spend all its time on the numerical application. Lastly, the 1 / 0 processors permit high-bandwidth 1/0 transactions To meet price-performance targets for a new minisupercomputer, a team of computer scientists conducted an exhaustive-and enlighteninginvestigation into the relative merits of available architectures. with minimal involvement from the interactive or numeric processors. A bersion of this article appeared In froc. 22nd Huwuii lnl'/ Con5 on Syslerns Sciences, Jan. 3-6. 1989, Kailua-Kona, Hawaii. work done at T R W Array Processors and at ESL (a subsidiary of TRW). The polycyclic architecture3 developed a t TRW/ESL is a precursor to the directeddataflow architecture developed at Cydrome starting in 1984. The common theme linking both efforts is the desire t o support the powerful and elegant dataflow model of computation with as simple a hardware platform as possible. The driving force behind the development of the Cydra 5 was the desire for increased performance over superminis on numerically intensive computations, but with the following constraint: The user should not have to discard the software, the set of algorithms, the training, or the techniques acquired over the years. As a result, the user would be able t o move up in performance from the supermini to the minisuper in a transparent fashion. This transparency is important for a product such as the Cydra 5 , which is aimed at the growth phase of the minisupercomputer market. Such a product must cater to a broader and less forgiving user group than the pioneers and early adopters who purchased first-generation minisupercom-12 00l8-9162/89/0l00-00I2$01 00 c 1989 IEEE COMPUTER
doi:10.1109/hicss.1989.47160 fatcat:binrbcntczd6le3rvklrtkqqm4