The 3B2 was designed for AT&T by Convergent Technologies.  I later worked with several people at Convergent, one of whom had a framed circuit board on his wall.  It was a wonder to behold -- the board had wires all over it that were added later, and nearly a dozen "bugs" -- in the days of discrete logic chips, a bug was when you took another chip and glued it, upside down, on top of an existing chip and then ran wires to the pins in the air.   As I recall, the story was that the first demo of the 3B2 happened roughly six weeks after the initial request, using the board on the wall.  Now, that's what should really be in the computer museums...

In those days, if there was floating point it was a separate chip, and the 3B2 had none.   Floating-point instructions caused a fault, which meant a context switch to the OS, where the instruction was emulated and then the program returned.   The performance, as I recall was about 800 FLOPS - dismal.   We fixed the compiler so it would generate calls to subroutines that did the floating point operations, and the performance improved by over an order of magnitude -- still dismal, but no longer ridiculous...

One of the events that led me to leave AT&T was that they fired the head of the benchmarking group at Indian Hill, a most competent woman, because they didn't like the results she was presenting.  When a company's information channels stop functioning reliably, it's time to leave...

Steve



----- Original Message -----
From:
"Doug McIlroy" <doug@cs.dartmouth.edu>

To:
<tuhs@tuhs.org>
Cc:

Sent:
Sat, 30 Jun 2018 14:24:24 -0400
Subject:
Re: [TUHS] AT&T Hardware


Anent 3B's: Last time I visited Paul Allen's Living Computer Museum
the only working Unix on display was running on a 3B2. Apparently
the machine was robust if nothing else.

doug