cthombor@luke.d.umn.edu (Clark Thomborson) (01/25/89)
...You didn't really think we'd all agree on the validity of Teller's claim, did you? You all have access to the postings to sci.military, which were mixed in their assessments. I also received some interesting technical info by direct email, summarized below. (I hope the authors don't mind...) Amos Shapir of National Semiconductor (IC) Ltd, Israel: I do believe today's technology can put a cray-1 in a card box - the NS32532 is 1" square and does 17000 dhrystones - almost as fast as a Cray-1! But keep in mind: 1. The Cray-1 is a rather old machine by now; and 2. We are talking scalar operation here, and the Cray's strong point is its vector capabilties. Phil Lapsley of UC Berkeley: [Teller] might be carrying around a miniaturized version of a project produced at Los Alamos a year or two ago, which was essentially a cube 6" on a side filled with 68020 CPUs in parallel. At the time the LANL folks planned to use the system for intelligent re-entry vehicles, the idea being that the RVs would have the smarts to avoid Soviet kinetic energy interceptors. They claimed their 6" cube was equivalent to a Cray I in speed. But a Cray I isn't that fast of a computer anymore. I'm still left with my original technical question, sharpened by all this discussion. Here are some specs for a Cray-1: 160 MFlops peak performance (64-bit floats) low vectorization penalty (speedup even on short vectors) 1 million words of memory at 320 gigawords/sec 12 I/O channels It's relatively easy to get 160 MFlops into a small box. I still don't believe the rest of a Cray-1 will fit there, certainly not for $10,000 in commercially-available parts. Right? And if we let Teller define `computing power' as ALU or even CPU data bandwidth, then his box reveals little about the cost or technical feasibility of putting a Cray-1 in space, to say nothing of making a Brilliant Pebble... Of course, the most important questions raised by Brilliant Pebbles or any other weapon system are the political ones. Still, it would be nice to get the technical facts straight. Clark