Debugging the card reader is tricky. The card reader is controlled by numerous relays, electromechanical switches that are prone to malfunctioning if their switch contacts get dirty. The photo shows rows of relays above a jumble of diodes and resistors. 2/8

@kenshirriff it seems strange to think that it wasn’t until the i486 (1989) that this line of CPUs had on-chip cache. But suppose I am comparing an 8088 CPU with 28k transistors to one with over 1.1m

@kenshirriff The 8 bit bus on the Intel 8088 made the IBM PC compatible with the peripherals available for existing 8 bit PCs. DEC even introduced a "Rainbow" with both an 8 bit CPU and an 8088 in the same PC, so it could run both old & new software.

@kenshirriff Thank you for these threads. They are always so fun and interesting to read.

A punch card holds 80 characters, one per column. The holes in each column represent the character. Cards can be punched manually with a typewriter-like keypunch. Or the computer can punch cards at high speed through an attached reader/punch. 2/9

@kenshirriff First parse of that had me reading the Twitter.com still used punch cards. And nothing would surprise me about Twitter.

When you push the POWER ON button, the computer springs to life. But starting last week, it would power off as soon as you take your finger off the button. Something was going wrong with the power-up sequence. 2/10

@kenshirriff I had the privilege to see the 1401 in action in the museum last October.

Thank you (and all volunteers) so much for keeping it running. 🙏

P.S. As a former Sun Microsystem employee, seeing only a few pieces was a pity, even if Sun was not far away.

The 1553 bus transmits data at 1 megahertz using 27-volt pulses on two wires. The chip receives bus data and converts it to digital (TTL) levels. The chip also transmits data on the bus, generating high-current pulses. 2/18

This chip was built by Integrated Device Technology (IDT) in 1991 for a military application. It was mounted on a multi-chip module with EEPROMs on either side. 2/13

Modern processors cache megabytes of data so it can be accessed without waiting on slow memory. The 8088's prefetch queue held just 4 bytes, still enough for a substantial performance boost. When memory was idle, the 8088 would load upcoming instructions into the queue. 2/12

@kenshirriff 6502!

Modern processors cache megabytes of data so it can be accessed without waiting on slow memory. The 8088's prefetch queue held just 4 bytes, still enough for a substantial performance boost. When memory was idle, the 8088 would load upcoming instructions into the queue. 2/12

@kenshirriff ...are those hand-tinned circuit boards?

@kenshirriff could there have been fume emissions concerns with other printing methods?

paging @plaidtron3000

Of course the Arma Micro Computer is not a microcomputer by the modern definition, since it was built from transistors. But it was small enough to get the name "Micro Computer". It also shows that 1960s computers weren't all room-filling mainframes. 2/13

@kenshirriff I would say the first personal computer has the Kenbak-1, but that's not quite what you asked :P.

@kenshirriff I would have said something in the PDP-8 family. Looks like the 8S had a desktop model (1966). But this beats the PDP-8 by a few years, and it's much smaller.

The cylindrical motor/generators rotated shafts inside the CADC, with a feedback loop to determine the right position. The tachometer generator provided speed feedback to the control circuitry, slowing the motor as it approached the right position instead of overshooting. 2/16

@kenshirriff do you know if this is the same Bendix as the Bemdix G-15 (?) that Usagi Electric is restoring?

@kenshirriff thanks 🥰🥰🥰

@kenshirriff Thank you for sharing this. I wonder if the reason for doing things this way was for radiation hardening.

The 386 was originally scheduled for 50 person-years of development time, but it fell behind schedule. The designers made a risky decision to use "automatic place and route", letting software do some layout. This worked and the chip was completed ahead of schedule. 2/17

@kenshirriff

Been pretty excited when i bought my 386-25Mhz Board/CPU for then ~1200DM in 1986...

@kenshirriff @nina_kali_nina Last x86 cpu successfully cloned in the SU!

@kenshirriff
My first 386 ran at 12MHz and addressed 640K RAM. I thought my setup was so quick.

@kenshirriff I remember connecting one :)

@kenshirriff I worked intimately with the design and analysis of these PGA packages in my early career then.

Interesting psychology anecdote – when you had a sample and you showed it to people (eg upper management) and told them explicitly do not touch the interior, they would find it irresistible and stick their finger into the inner die-attach surface and bonding pads anyway, ruining the part. It was very weird how they couldn't control themselves. Must. Jab. Finger. In. Package.

Here's an X-ray of the 386 package. The pins are bright white. The gray square in the middle is the silicon die. Inside the package, the grayish traces connect the pins to the contacts surrounding the die. Thin bond wires connect these contacts to the die. 2/13

@kenshirriff tbh i can't see anything other than the bent pin, as i find those deeply upsetting lol

Since the chip is CMOS, it contains two complementary types of transistors—NMOS and PMOS—working together. They are built by doping silicon to form N and P regions. The NMOS transistors are constructed in a "well" of P-type silicon. 2/20

Each RAM cell consists of two inverters in a loop, a standard design. Two pass transistors select the cell for reading or writing. This cell has 6 transistors so it is known as a 6T cell. 2/8

An aircraft has a pressure port to measure the static air pressure and a pitot tube to measure the total pressure due to speed. These were connected by tubes to the front of the CADC. 2/13

@kenshirriff so it's just a blowtorch with a slide rule spackled on to it?

This "alloy junction" transistor has a germanium wafer with a thick indium ring on the visible side and an indium disk underneath. The indium forms an alloy of P-type germanium. The ring is the emitter, the disk is the collector. The result is a PNP transistor. 2/8

@kenshirriff Omg, this is so beautiful. It would be a glorious poster!

@kenshirriff Fascinating ! 2 questions, if I may: 1) is that some type of input protection? 2) the datasheet states HDB3 outputs can source 10mA, whereas all other output currents are 1.6mA. So I went looking for larger drivers, but all outputs look similar. Am I missing something (maybe it's just that the design got revised?). Thanks for all those nerdily interesting posts 🙂

@kenshirriff common strategy for rad-hardening