Comments by "Lawrence D’Oliveiro" (@lawrencedoliveiro9104) on "" video.

  1. 11:49 The legendary Seymour Cray had a simple solution to the problem of divide operations taking longer than multiplies in his supercomputer designs: he didn’t have a divide instruction at all. Instead, you had to multiply by the reciprocal of the divisor. He provided a special instruction for computing the reciprocal; the first execution of this instruction produced an approximate result which was lacking accuracy in the last few bits, so you had to execute it a second time on the output from the first time, to get the fully-accurate result. The assumption, I guess, was that you would tend to do a lot of divisions by the same divisor. Also you could skip the second iteration if you didn’t need the full accuracy in the result.
    3
  2.  @BurnleyNuts  I taught my elderly mum how to program her own video recorder, back in the day. I never understood all those jokes about how hard it was.
    2
  3. They were already going out of fashion when I started my Comp Sci degree in 1979. I’m not really sure they help with program comprehensibility or reliability in any way. Look at 15:25, for example, those arctangent functions: even by that time, FORTRAN had in its maths library a function called ATAN2, with separate X and Y arguments (instead of passing their combined ratio) so you could use their respective signs to uniquely determine the result angle over a full circle. That is missing here. Just because it looks pretty, doesn’t mean it’s actually correct.
    2
  4.  @SeanBZA  When did you do that? I look at something like SiFive, and they claim they can put together a chip implementation for you within weeks.
    2
  5.  @jessepollard7132  From the Cray-1 Hardware Reference Manual (published 1977), page 3-28: “The division algorithm that computes S₁/S₂ to full precision requires four operations: 1. S₃ = 1 / S₂ -- Reciprocal approximation 2. S₄ = (2 - S₃ * S₂) -- Reciprocal iteration 3. S₅ = S₁ * S₃ -- Numerator * approximation 4. S₆ = S₄ * S₅ -- Half-precision quotient * correction factor The approximation is based on Newton’s method. The reciprocal approximation at step 1 is correct to 30 bits. The additional Newton iteration at step 2 increases this accuracy to 47 bits.”
    2
  6.  @StevePemberton2  If you’re interested, look up the Friden calculators from the 1940s/50s. They could do multiplications and divisions by repeated addition/subtraction. Later models were electrically-driven.
    2
  7. Years of training, and I think also they had to follow written checklists, and of course Mission Control was always on the line. 90% of their training was how to deal with things going wrong.
    2
  8. I heard the inside of the Apollo capsule described as similar in size to a “station wagon”. So nothing from NASA (from that era) could be described as an “apartment building”, by any stretch. Except maybe the inside of Skylab ... Why the Americans never adopted a Soyuz-style design (like the Chinese did), with more room available to the crew in space while still weighing less overall, I’ll never understand ... Trivia question: why did Gemini have ejection seats? Both Mercury and Apollo had that abort rocket on top of the capsule, to get the crew away in case of a booster failure, while Gemini did not. Why?
    2
  9. Those delay lines would only have been needed on the French SECAM system. The Germans came up with a much simpler solution to NTSC’s colour problems (besides its inability to spell “colour”).
    2
  10. That was the one part of the Apollo program with probably the most women involved.
    2
  11. SIMH does all the PDPs, I would say. (Except the oddball ones which weren’t really computers.)
    2
  12. It was a product of its era: Knuth was trying to be as architecture-agnostic as he could, because “decimal” computers were still a thing. I think in the later editions he threw all that rigmarole away.
    2
  13. 15:45 The IBM computer that controlled the Saturn V was a hybrid digital/analog design: the digital part wasn’t fast enough to make corrections to thrust magnitude/direction etc in real time, so that was left to the analog part, while the digital part just computed the direction the rocket needed to go. (I think it could only come up with new numbers once a second, or something of that order.)
    1
  14. The old DEC machines (from the same era) had that “page” bit (to select either “same page as current instruction” or “page 0”) but they also had an “indirect” bit, which meant that the word being addressed could hold in its turn a full indirect address. That seems to be missing here.
    1
  15. There were quite a variety of word lengths in the days before byte-addressability became the standard. DEC’s first machine, the PDP-1 had an 18-bit word length; their very popular PDP-8 range used 12 bits (awkward to program, but cheap for those days); and their big “mainframe-class” PDP-10 had 36-bit words. I think the CDC 6000-series supercomputers, designed by the legendary Seymour Cray, had a 60-bit word length. The guys at AT&T Bell Labs who invented Unix did their first prototype on an old PDP-7, another of DEC’s 18-bit machines, that they had lying around. That proof of concept allowed them to get the funding for a nice new, 16-bit byte-addressable PDP-11. And the rest, as they say, is history.
    1
  16.  @jessepollard7132  ECC memory can be done with extra checking bits that are only visible to the hardware, not the software.
    1
  17.  @Hackanhacker  Hard to see how it is supposed to cope with revisions to the program code. Say somebody sends you a patch in the usual “diff -u” format: how are you supposed to apply that to the flowchart?
    1
  18. 15:04 My Python-based desk calculator tells me that .406736643 is the sin of 24°, accurate to about 10 figures.
    1
  19. We still have custom computer chip designs. You can put together a bunch of building blocks (e.g. a RISC-V CPU core, some RAM, some EPROM, some interrupt lines, maybe a display controller or Ethernet or Wi-Fi or USB or whatever), and a fab will do a run of chips for you. There might be a minimum number of a few thousand units or something like that.
    1
  20. 15:14 Even using the same forms as for part drawings. Did you notice the section at the bottom right, to do with measurement tolerances etc, which was crossed out?
    1
  21. Tape backups still exist, and are still being developed. The magnetic technologies that allow us to cram more and more data on hard disks can also mostly work for magnetic tapes.
    1
  22. They were ROM implemented as core memory. Whereas read/write cores could have their magnetization changed and sensed, these stored bits based on the presence or absence of individual wires. The piles of wires looked like ropes, hence the name, I guess.
    1
  23. 1:15 Fun fact: the first Apple Ipods (remember them?) had little Toshiba hard drives in them. Yup, spinning drives (I think 1¼” diameter) in a device that would be moving around a lot, kept in people’s pockets or purses, likely dropped and bumped a few times. And still keep working, without a bunch of engineers back at Mission Control continuously monitoring things. The market window for these tiny drives was brief, before flash memory completely took over from them.
    1
  24. There would be non-invasive ways of reading out the memory state, with external passive probes or something.
    1
  25. Remember to pull the crank up, not push it down.
    1
  26. They had multiple big IBM computers back at Mission Control. This on-board computer was really just interpreting sensor readings, little more.
    1
  27. Apple manages to design its computers to prevent users from doing things (like running alternative browsers). That sucks up a lot of computing power, too.
    1
  28. His sense of UI design was also a factor in the death of John Denver.
    1