Warning: arcane big iron talk. Attack of the Killer Micros versus Hairy Smoking Golfballs
Great summary Tim, about supercomputers. Still, I think that the majority of them in history that actually went into production from CDC, Cray Research, NEC, IBM, Fujitsu, Cray Computer, etc., were proprietary, designed from the ground up around custom gate arrays. I did see some of Intel's 9000+ Pentium Pro machine (for Sandia Labs?) being put together up in an Intel Oregon facility about seven years ago. I recall that the interconnect and packaging were extremely well done. Impressed me so much for a "chip company" that I think I bought some stock! (actually, seeing the first Intel PPro servers at about the same time, 2 way and 4 way, probably had more to do with it). I think Adam Latham worked on that Sandia machine, on the software side.
-- the very hard to cool processors of the IBM mainframes...IBM had to develop the sophisticated "thermoconduction modules" with plungers and whatnot to cool the CPUs of their bipolar machines
Yes, water cooled machines were it in mainframes from the 70s to about 1990. As you point out CMOS just wasn't ready for big machines, for quite a few years of its existence, because it was too slow (actually very weak drive capability) and had terrible skew problems across multiple drivers. Those were all fixed (I'm sure Intel had a lot to do with it) and big iron crossed over to CMOS in the early 90s.
what I think of as the "ultimate last gasp" for bipolar or ECL processors was Trilogy. Doug Peltzer set out to built a very high speed mainframe using bipolar and ECL technologies. They failed, and Trilogy folded.
At Trilogy, Gene Amdahl tried something just too far out: wafer scale integration he called it. Actually tried to make chips the size of wafers, maybe 5" at the time. Of course (we can say now) the odds of getting an "all good" "chip" were about as impossible as they would be today. Actually, it wasn't as impossible as it sounds, as they had lots of redundancy on the wafers, and planned to use discretionary wiring to hook up the good parts, but still, no dice (no pun intended). Well, the discretionary wiring on a piece of silicon 5" in size probably was nigh onto impossible also.
That wasn't the last gasp for bipolar mainframes, though. IBM, Amdahl Corp, and Hitachi went on to do a couple of more generation in the next 6 or 7 years. Just for history, the densest ECL, water cooled chips, might have been 20 - 30K gates (not including RAM, just logic). Far cry from CMOS, which was probably already at a million gates/chip in 1992?
Good history diversion, Tim.
Tony |
