Build your own Apollo Guidance Computer

William Donzelli wrote: Adding just ten more chip types will not significantly increase manufacturing overhead. How many parts total were on the Cray-1 BOM? 10,000? 20,000? making it 10,010 or 20,010 is not going to strain the inventory folks. Also, by adding just ten more chip type, the total number of chips in the system will have gone down - perhaps by ten percent. That will greatly increase yield. Back in those hand-placed hand-soldered surface mount days, board errors were far more likely to be bad solder joints (using that old fashioned solder! horrors!) than misplaced parts. One of the Cray-1s faults was the somewhat miserable servicing requirements, and the crummy downtime that forced it. Fewer parts would have increased uptime (although many of the issues with the machines were not chip related). -- Will -------------------------- I think a bigger factor here was the consistentcy of switching time. All of Seymour's design are incredibly tight on switching time. He liked to line up signals so they would reach the next gate at the same time without requiring a clock. Having only one gate switch time would ease design. In all of his designs that I have worked on, he made flip flops and adders out of individual gates. Again to have exact control of the switching time. And to eliminate any un-necessary logic. There are no unused gates in any of his computers. So I don't think inventory count was a factor. It was just his design phiilosophy - minimum logic, exact timing. There is a great book on this: "Design of a Computer: 6600". Al has it on his site; well worth downloading and reading. Miserable servicing? As someone who spent literally years tuning wires in Seymour's designs, I have to agree. His machines were very demanding to impossible to maintain. Just before it died, I spent a few hours on the 8600. None of us on that machine believed it could be maintained! And the math models all gave the MTBF as a negative number! Billy