[HN Gopher] Oral History of Robert P. Colwell - Chief Architect ...
       ___________________________________________________________________
        
       Oral History of Robert P. Colwell - Chief Architect for Intel's
       IA32 [pdf]
        
       Author : areoform
       Score  : 48 points
       Date   : 2023-11-29 13:34 UTC (9 hours ago)
        
 (HTM) web link (www.sigmicro.org)
 (TXT) w3m dump (www.sigmicro.org)
        
       | areoform wrote:
       | Some of the stories that Colwell shares are eye opening. BC in
       | this part of the transcript stands for Bob Colwell,
       | 1:15:56 BC: Yeah. So I've got these notebooks all the way back to
       | the seventies. That's just how I organize my head and I can find
       | things that I want.                   I have lots of pictures
       | from my battles with this problem at Multiflow tracking this
       | down: it has got to be this chip, it is not my design, it has got
       | to be the chip. Eventually, Rich Lethin and I made a pilgrimage
       | down to TI in Richardson, Texas and we said as best as well can
       | tell many of your chips don't work properly. And does this come
       | as a surprise to you?"                   I half expected them to
       | say, "what you are out of your mind! You've done something wrong.
       | Come on, you don't know what you're doing. Go use somebody else's
       | chips."                  But no, they said "Yeah we know, let me
       | see your list." And they looked at the list and said "here is
       | some more that you don't know about."         [laughter]
       | 1:16:39 BC: And we went "Thank you very much. This is what we
       | needed." At lunch I asked them "how did this happen?" and by the
       | way it wasn't just TI. Their parts were no worse than anyone
       | else. Motorola's were no good, Fairchild's were no good, they all
       | had this problem.                   And so I asked TI, "how did
       | the entire industry fall on its face at the same time? We are
       | killing ourselves trying to work around the shortcomings in your
       | silicon."                  And the guy said, "the first
       | generation of TTL was done by the old gray beard guys that really
       | know what they are doing. The new generation was done by kids who
       | are straight out of school who didn't know to ask what the change
       | in packaging would do to inductive spikes."                    He
       | pointed out that what had happened was we had gone from 14 pin
       | DIPs, which are small, to 24 pin DIPs which are really long and
       | he said the dies right in the middle of that package and the
       | power and ground have long leads, from the package out to the
       | corner pins. Now if that had used middle pins they would have
       | been short. But at the corners they got really long and therefore
       | power/ground current inductance got really bad.
       | I said "that is just wonderful because we picked TTL specifically
       | to try to avoid any such problems with the implementation
       | technology." We said to ourselves, Multiflow machines are going
       | to be such a risk at an architectural level and should pay off so
       | well that we can back off the aggressiveness of the silicon
       | technology, we will make up with it for the compiler. We will
       | take no risks here.                  Instead we signed up for
       | this TTL family, and inadvertently just shot ourselves in the
       | foot with it. It took us about a year to work around all the
       | different ways in which it was failing. It was incredible. You
       | learn a lot that way.
        
         | bigbillheck wrote:
         | > The new generation was done by kids who are straight out of
         | school who didn't know to ask
         | 
         | I wonder how many of those kids would, thirty years later, go
         | on to complain about kids these days.
        
           | AnimalMuppet wrote:
           | The kids in question, thirty years later, _did_ know to ask.
           | And the thirty-years-later-just-out-of-school kids _didn 't_.
        
           | Exoristos wrote:
           | Because no one and nothing is ever better or worse than
           | others, right?
        
       | Torkel wrote:
       | Bob Colwell is awesome! I remember seeing an online talk he
       | made... must be 25 years ago now. He talked about the development
       | of the out of order architecture at Intel, and in general about
       | engineering. He has some really fun anecdotes!
       | 
       | Honestly he is probably a big reason I switched majors from
       | biotech over to computers, robotics and engineering in general...
       | I even remember sending him an email about it and got a response
       | back, wishing me good luck :) Sadly I can't find the email now,
       | it was pre-gmail.
       | 
       | Some really good columns that he wrote for "Computer" magazine:
       | https://people.computing.clemson.edu/~mark/330/colwell/at_ra...
       | 
       | Highly recommend his book "Pentium Chronicles"
       | 
       | Edit: found the talk! It was ee380 at Stanford 2003/2004. Sadly
       | the video links seem to be broken...
       | https://web.stanford.edu/class/ee380/ay0304.html
        
         | KerrAvon wrote:
         | Second "Pentium Chronicles." It can be a slog in parts if
         | you're not _really_ into silicon validation, but the anecdotes
         | are worth it.
        
         | ghaff wrote:
         | He was also at DARPA. There a presentation deck floating around
         | from maybe 10 years ago--maybe HotChips--that, among other
         | things, comes across to me as a rather pessimistic take on the
         | petering out of CMOS process shrinks aka Moore's Law. His take
         | was basically that, sure there are a bunch of other things you
         | can do, but CMOS process shrinks have really been a unique
         | thing in terms of achieving performance increases in the
         | computer industry against which all else pales.
        
           | lswswein wrote:
           | I believe this is the talk from hotchips:
           | https://www.hpcwire.com/2014/05/28/planning-post-moores-law/
        
       | groos wrote:
       | Bob Colwell used to write a column for IEEE Computer (IIRC) that
       | was just awesome reading.
        
       ___________________________________________________________________
       (page generated 2023-11-29 23:01 UTC)