The University of Iowa's DEC PDP-8

Restoration Log

Part of the UI-8 pages
by Douglas W. Jones
THE UNIVERSITY OF IOWA Department of Computer Science



This is a chronological log of the progress restoring the University of Iowa's PDP-8 computer. Entries are added at the end as work progresses. Click on any thumbnail image to see full-sized image.

Jan. 13, 2015, Teletype Print Hammer

  thumbnail image     thumbnail image  
Nov. 2013Jan. 2015
  thumbnail image     thumbnail image  
Replacing the print hammer
Bug 9: In the year since we began work on the PDP-8, several people have used their thumbnails to "test" the state of decay of the print hammer. In November 2013, the hammer looked reasonably good but had a badly worn face. By January 2015, despite no use for printing, the hammer had been significantly reduced by "testing." This wear is clearly evident in first two photos shown here.

Dan Tumey has invested in sinking a mold for Teletype print hammer heads, and he sent us a baggie of heads. Cleaning off the old head was messy; it came off as crumbs (I collected them for a photo), and the sticky residue was difficult to remove from the metal body of the hammer. Careful scraping was required to get down through the residue to the metal hammer body, and the groove that retains the hammer head was especially difficult to clean. I scraped out what I could with a watchmaker's screwdriver, and then used a loop of Goo Gone soaked cotton string to erode out most of what was left. The final step in cleaning the groove was done with 5 turns of dental floss.

A drop of silicone oil (the recommended lubricant for the clutch pads on our Tally high-speed paper-tape reader) was all it took to allow the new hammer head to pop into place.

Bug 23: As noted on Feb. 11, 2014 there were only 3 bolts out of the 4 required to properly mount the Teletype sub-base. Inspecting these bolts, no two of them were the same, and none were exact matches for the two options mentioned in the Teletype parts list. The parts list indicated that 10-32 screws and had been used with an early version of the sub-base casting, while later production used a self-tappng screw and spring clips. We found one such spring clip among the loose debris inside the Teletype case. Having a number of 10-32 rack mounting screws, I cut 4 of them to the correct length and measured the pockets in the sub-base that are designed to hold the spring clips. These pockets are 1/2 inch by 3/8 inch; we will make rectangular 10-32 nuts to fit.

Jan. 14, 2015, Teletype Base Assembly

  thumbnail image  
Teletype base
Bug 23: I made 4 10/32 nuts out of 3/8 inch wide chunks of 1/2 by 1/8 inch steel bar, fitting each into its pocket in the Teletype base casting. This required a bit of filing -- the steel bar was a bit oversize on the 1/2 inch dimension. With the nuts sitting snugly in their pockets, I set the casting on top of the stand and screwed in the screws from below, with an oversized washer on each screw.

Bug 21: Dan Tumey sent a new foam pad along with the new print hammers installed on Jan. 13. This fits nicely, as can be seen in the photo of the assembled base and stand.

Bug 26: I cleaned the paper-tape-punch base casting and attached it to the side of the main Teletype base casting, leaving the miscellaneous unidentified hardware that was found in the Teletype in baggies with labels in the paper-tape-punch base. There is plenty of room there. I also re-attached the paper-tape reader subassembly to the main casting and did a preliminary cable routing. The keyboard cable needs to be routed along a parallel path before the Call Control Unit is replaced.

Jan. 22, 2015, TTY Keyboard, PDP-8 Backplane

Bug 41: In order to solve Bug 27, we will need to further disassemble the keyboard in order to remove the key caps from the mechanism. This is risky because putting it back together will be a challenge. In investiganting how this might be done, I discovered that the keyboard cover (the part with one hole per key) is missing three mounting lugs that align it and support it on the metal brackets at each end of the keyboard. The pins were 3/16" diameter plastic protrusions at least 1/4 inch long, judged by measurements of the holes into which they fit and by measurements of the scars left behind when they broke off.

Bug 42: In studying the wire routing on the cleaned and reassembled Teletype base, we looked again at the keyboard contact block assembly and its trailing wiring harness. The contact block is clear plastic (yellowed with age), with a strip of metal contacts (separated by insulating layers) across the top that is held on by spring clips on 4 lugs that stick up from the plastic body. Two of these lugs have broken off (and one spring clip has gone missing). Having corresponded with a supplier of Teletype parts, we found that we can get a new-old-stock contact block, so we will rebuild our contact block assembly.

Bug 19: Since Oct. 2, 2014, we have been working on reverse engineering the wiring of the backplane around slots PA29, PA36, and PB22. Zimu Zhang and Bowen Bai traced the connectons to these slots and found much of the relevant documentation. This work made it clear that slots PA29 and PB32 are wired in conformance with the maintenance manual, and that the wiring is in clear conflict with the boards in those slots. In addition, markings on the handle of the board in slot PC33 strongly suggested that it had originally been in slot PA29 -- the neighboring boards in row PA all had similar marks. Therefore, I made the following changes:

SlotDocumentedAs FoundAs Changed
PA29 R603 R1115 R603-x
PB22 R111 R603 R1115
PC33 R603 R603-x R603

Notes: The notation R603x indicates a board with a marked handle. In addition, it appears that the R1115 board is a revised design of the R111.

This leaves one discrepancy to be resolved: Slot PA36 still contains an R302 board where the documentation leads us to expect a W501.

Jan. 23, 2015, TTY Keyboard, PDP-8 Rack Wiring

  thumbnail image     thumbnail image  
Teletype keyboard cover
Bug 41: We drilled 6 pilot holes 1/16" diameter in the teletype keyboard cover, each carefully centered in the scar of one of the broken lugs that formerly held the cover in position. Each hole was then drilled out to 5/32" and then tapped for 10-32 threads; ABS plastic taps easily. After tapping, I found that 10-32 truss-head screws, as commonly used to mount rack-mounted hardware, would fit, with the screw heads just clearing the plastic Teletype case.

For the 4 corners of the keyboard cover, I screwed in these screws from the inside. After we learn the correct measurements for the mounting lugs, we will cut these screws to length. For the 2 central screws on each end of the keyboard cover, I screwed in the screws from outside so that the screw heads can serve the purpose formerly served by the C-clips on the central lugs. The left photo shows the test fitting of these screws in one end of the keyboard.

Because there may be some tension on these screws, I backed each of them with a nut, gluing bits of ABS rod on the inside of the cover next to each nut to serve as a wrench to prevent the nut from turning. This is probably overkill, since the amount of tension on these screws should be minimal. The right photo shows the result as it was test-fitted on the keyboard. The test showed that there is ample clearance between the screws extending inside the keyboard and the keys, although the Break key comes close enough to the center screw on the right end that we will probably cut that screw a bit shorter.

A second repair to the Teletype keyboard cover is visible in the photos: There are pockets on the upper left and right corners of the keyboard cover that mate with corresponding tabs on the Teltype cover. The outer side of the pocket at the upper right corner of the keyboard was missing, broken off at some past time. I made a replacement from scrap ABS and glued it in place with Plastruct Bondene solvent cement, and then reinforced the joint with smaller ABS scraps. Color was ignored in this repair because, with the main cover on, it is invisible.

Bug 16: We began work trying to understand the role of the Type 834 power control panel. We have the schematic for the panel, but we need to understand its use in context. Initial efforts suggest that the full current to power the PDP-8 is passing through the front panel key switch, while the power control panel is only controlling power to the ADC rack. This does not seem right, given that all of the 110V wiring in the power supply is 14 gauge, while the wiring to the key switch is 18 gauge. Would it make more sense for the key switch to control the contactor in the power control panel, and have that contactor control the power to the entire system?

Jan. 27, 2015, PDP-8 Rack Wiring and Backplane

Bug 19: We have completed analysis of the wiring slot PA36. In the maintenance manual, the Processor Utilization Module List in the PDP-8 maintenance manual clearly show a W501 board (a Schmitt Trigger) in this position, but our machine has an R302 board (a dual one-shot).

After mapping all of the wiring to this board slot, we determined that the wiring is incompatable with a W501 and entirely compatable with an R302. Furthermore, close inspection of the backplane shows that, while most of the backplane wiring was done with an Automated wire-wrap machine, as evidenced by very clean horizontal, vertical and diagonal wires and a uniform quality of wrap, all of the wires to slot PA36 are were randomly routed and hand wrapped. Therefore, we conclude that the R302 is the correct board for this slot, and that this slot represents a field wiring change made after the machine was manufactured.

We should try to document the change that was made and relate this to the logic diagrams for the CPU, but it should be safe to power up the machine as is. The logic diagram for the board is given in the Maintenance manual, in the drawing of the Timing, Keys, Switches, and Run Control (coordinates A2 in the upper left of the drawing). This appears to be switch debouncing logic for several of the front-panel keys.

  thumbnail image  
the Mercury contactor
Bug 16: The most prominent component of the power control panel is a mercury contactor. This is a normally open device with the following specifications on the label wrapped around the coil:

EM-1 115 V.
   60 cy.
115 V./60 A.
230 V./35 A.
120 V./12 A.
220 V./ 7 A.
A.C. - 2 HP.
D.C. - 1/2 HP.

I tested it, and it worked. When on, the pool of mercury rises far enough to be visible inside the sealed glass envelope. Note that, if anyone ever sends this machine to the landfill, the mercury relay should be removed so that the mercury can be safely recycled.

Bug 41: Wayne Durkee confirmed (by E-mail) that the correct length for the corner lugs on the sides of the Teletype keyboard is 1/4 inch, so we cut the screws we had installed there to that length, or just slightly longer.

Jan. 29, 2015, Teletype key replacement

  thumbnail image     thumbnail image  
Pulling a keyReplacing it
  thumbnail image     thumbnail image  
Rebuilt keyThe result
Bug 27: We went over the keyboard deciding which keys we wanted to replace, and for each, we found its part number in Teletype Bulletin 1184B (the parts list). We have a Nov. 1964 and a March 1965 edition, and the March 1973 edition is available on line; unfortunately, the lists of keys in these editions are not the same. Paul Cembura, doing business RTTY Elctronics of El Sobrante, California under the nickname Mr RTTY stocks a variety of "new old stock" Teletype parts for retail sale; he had exact replacements for most of our bad keys, and reasonable alternatives for the others, so we ordered replacements from him, along with a new ribbon and a new keyboard contact block (see Bug 42, below). Unfortunately, we could not get replacements for the Control and Shift keys, but while those were cracked, they were better shape than the other cracked keys.

Replacing the keyboard keys proved a challenge. Wayne Durkee, who had previously provided some Teletype parts, sent us helpful instructions. Pressing any key down, with the keyboard cover removed, allows that key assembly to be tilted up and unhooked from the keyboard frame (after pressing any key, you need to depress the keyboard locking lever on the lower front right of the keyboard frame to reset it and allow the next key to be pressed).

As seen in the photo a rebuilt key, each key assembly consits of a two-part injection molded keycap press-fitted ont to the up-facing stem of a stamped steel "keylever" (in classical typewriters and earlier model Teletypes, the corresponding part was an actual lever; in the model 33, it is not). The keylever has a horizontal base that presses down on the code-bars in order to encode that key, with a retaining hook on one end and a guide prong that pokes down ward and rests on the key return spring.

To remove an old keycap, pull straight down on the keylever by the base of the stem, avoiding any off-axis forces that might bend the keylever. This is harder than it sounds. Doug Ives took a photo of me trying and failing to pull a key. Alternately, you can smash the key by hammering it on its side (being careful not to apply any force to the keylever). We did this with the key shown in the rebuilt key photo before we discovered a better way: Hold the key vertically, with the keylever stem poking up between the loose jaws of a pair of pliers held horizontally with the tips of the jaws resting on the edge of a table. Grasp the base of the keylever stem with a second pair of pliers held exactly vertically, in line with the stem. Press up so the keylever rises an inch above the horizontal pliers and then jerk down hard. The keylever will usually pop out of the keycap and the keycap will fly across the room, usually largely undamaged.

To install a new keycap, press-fit the keycap onto the stem. Again, this is hard. Set the keycap face down on a table, and press the stem in from above, using a pair of slip-joint pliers to grasp the key stem loosely, with the sides of the jaws of the pliers resting on the ears of the key stem. Then pound on the pliers to seat the stem in the key cap. The photo shows Doug Ives doing this.

The final photo shows the result, along with the old keycaps, some in fragments. The new keys are darker and shiny, while the old keys have a frosted texture as a result of age. It would be nice to polish them up, but this may not be possible. We tried solvent polishing on one of the old keys, and it made no appreciable difference.

Bug 26: Bug 42: As mentioned above, the new keyboard contact block arrived with the new key caps, so after the key caps were replaced, we began to figure out how to move the keyboard contacts from the old contact block to the new one.

Jan. 30, 2015, Rack Power Wiring

Bug 16: Here is the wiring diagram for the 110V wiring in the PDP-8 rack, the result of tracing out the wiring in the Type 834 Power Control Panel and the Type 708 Power Supply. This diagram describes the actual wiring of the machine, giving jumpering and connections to the power control panel that are not documented in the schematics for that panel, and including significant differences in the power entry to the Type 708 supply from what is documented in the schematics.

schematic diagram

The wiring of the white-white twisted pairs to the front panel key switch and the control for the contactor K1 in the Type 834 panel is clearly wrong, verging on dangerous. Depending on how the white-white twisted pair is plugged into the white-white faston connector pair on the back of the Type 708 supply, one of the following two circuits will result:
schematic diagram schematic diagram
Bad (as used)  Worse (a likely error) 

The way this was wired is safe in the most elementary sense: It does not blow up or short circuit, and the system will operate. In fact, it did operate, for a decade. On the other hand, this is not an approved way to wire circuits involving circuit breakers. If either of the two breakers is off, you would expect it to isolate all downstream wiring from the wall outlet, but as wired, because of cross connections, this is not true. If either breaker is on, the neutral wiring is all connected to the wall outlet. Normally, neutral is at a low potential, but miswiring in the wall outlet is not unheard of, and in some cases, this can put the neutral wire at a considerable potential. Furthermore, if the breaker in the Type 708 supply is off but the breaker in the type 834 supply is on, the neutral wires in the PDP-8 rack will be hot, energized through the K1 coil. In this state, they cannot source much curent, but they can source enough to be dangerous.

If the two white wires in the twisted pair are cross exchanged, the situation is worse. The coil for contactor K1 and the power to the PDP-8 are both stuck in the on condition regardless of the setting of the power key. The key only serves to short the two neutral lines. Furthermore, with the key open, the power to the PDP-8 is split between the two breakes.

In sum, had DEC asked for UL approval of this system configuration, it would have been denied. We can speculate about how this happened. The PDP-8 schematics we have from DEC are for a desktop machine. The power control panel is a standard part DEC was already using. It is clear that each of these can be used in confgurations that do not violate safety rules. It is fair to speculate that when the PDP-8 was configured as a large rack-mount system, technicians were left to modify and interconnect the components without adequate supervision. They produced an interconnection that worked. By 1970, the systems DEC was producing had significantly more systematic power control schemes, with a power-control bus connecting all cabinets. At some point after our machine was made, DEC learned how to do it right.

The good news is, the ADC rack, with the supply fully loaded, appears to draw a maximum of 440W. The PDP-8 is documented as drawing 750W in its basic configuration. We have the EAE option, which is an addition to the basic configuration, so our might draw 800W. Adding these, we get on the order of 1250W. Assuming a power factor of 1, that's only 11A from the wall outlet. We should be safely under the 20A capacity of the breaker in the power control panel, and therefore, we should be able to safely rewire the system to come much closer to a reasonably safe system.

  thumbnail image     thumbnail image  
Old and new contact blocks
Bug 42: Doug Ives set to work moving the wires and contact fingers from the old and yellowed contact block to the new block. Good photographs were crucial to this venture, in order to assure that each wire went to the same spot on the new block as it had gone on the old one. These are two of the photos he took.

Feb. 2, 2015, Teletype Rebuild

  thumbnail image     thumbnail image  
 Contact block Keyboard replaced
Bug 26: With the keyboard contact block reassembled, I mounted the contact block back into the right end of the keyboard. This is a bit tricky, as the contact block is held in place by two springs that fit in the round pockets at each end of the bracket, pressing it into place. Once it was assembled, we mounted the keyboard back to the Teletype base and clipped the keyboard cable into its path in the base casting.

Bug 22: With the keyboard in place, we reattached the call control unit and then "dropped" the typing unit into place, having decided to ignore the decay to the vibration isolators on which it rests. They are adequate in their current state. The tricky part of installing the typing unit is the H-shaped bracket that connects the keyboard locking lever to its actuator. This bracket sits just above the upper right corner of the keyboard, deep down in the works. Fortunately, the folks who designed the Teletype thought of this issue and added a slot to the bracked that just fits a screwdriver blade, so that a long screwdriver can be latched into the bracket and can be used to "steer" the bracked into place as the typing unit is dropped into place.

Feb. 4, 2015, TTY rebuild, Rack power wiring

  thumbnail image     thumbnail image  
 Teletype wiring Reassembly
Bug 24 and Bug 26: We cut a pair of 8-32 screws to size to replace the missing Call Control Unit mounting screws -- fillister head instead of the flanged hex-head screws used by Teletype corp. With the call control unit properly mounted, we then plugged in all of the plugs that fit the back of the unit. This was a puzzle; doing it right, many of the wire bundles help retain other wire bundles, with the net result that all of the wiring is neat. The other puzzle was the routing of the cable to the power supply we added to the Call Contol Unit. That cable fit nicely tucked down into the right side of the Teletype base, as can be seen in the photos.

Bug 16: Here is the wiring diagram for the revised 110V wiring in the PDP-8 rack connecting the Type 834 Power Control Panel and the Type 708 Power Supply. The line cord now enters at the power control panel, with the key switch connected solely to the contactor in that panel. The wires that formerly carried power from the Type 708 power supply rearward to the power control panel now carry power the opposite direction, and the lugs on the back of the Type 708 power supply that were formerly switched by the key switch are now bridged by a jumper.

schematic diagram

  thumbnail image     thumbnail image  
Type 834
Type 704
Moving the power cord to the back of the machine was easy; it was moved from the lugs on one circuit breaker to identical lugs on another. The nearby mounting screw for the input filter capacitor was actually more appropriate than the mounting screw the ground wire had been attached to. The power cord is very bulky and, as a result, inflexible, so it has to swing in large radius curves. We screwed a cord retaining clip to a rack panel mounting screw on the hinge side of the rack just out of view in the left photo shown here.

Moving the wire between the Type 834 panel and the Type 704 supply involved cutting off the ring-lugs that formerly terminated it to the input circuit breaker and replacing them with Faston connectors. We left the full length of this wire intact, so it can be swung over to the faston connectors on the input side of the Type 834 panel if someone wants to restore the machine to the "as used but unsafe" wiring that we have changed.

Because the Faston connectors that link the wires to the contactor in the Type 834 panel to the wires from the PDP-8 key switch are no longer plugged into the back of the Type 704 supply, we need to insulate them. We did this by plugging them into unused shrouded Faston connectors in order to completely cover the exposed lug, but this leaves a small area vulnerable. These connectors carry full line voltage, so we added some shrink tubing to completely shield them. We shrunk this tubing so that it is firmly attached while sill allowing the covered connector to be uncovered should someone want to restore the original machine configuration.

We added unused shrouded Faston connectors to the end of the power cable that goes to the ADC rack, covering both the line-voltage connectors and the logic power connectors.

After making these changes, we turned off the breaker on the Type 704 supply in the PDP-8 and then plugged in the machine, testing that the contactor in the Type 834 panel closes when the key is turned, and that the fan in the bottom of the rack and the pilot light on the Type 704 supply both come on when the contactor closes. This demonstrates that our rewiring is good.

  thumbnail image     thumbnail image  
As foundfixed
Inspecting the back side of the PDP-8 front panel, we noted a wire to nowhere hanging from the third terminal on the key switch. This wire was simply tucked up behind other wires, with no obvious destination, so we unsoldered it from the back of the key switch.

The extra wire had a bit of shrink tubing protecting its connection to the key switch, but there was no protection on the power lines. Furthermore, all of the connections to the key switch were simple lap joints to lugs just 1/8" long -- hardly an appropriate connection for the 7 or so amps that formerly flowed through the key switch! We unsoldered these wires, added shrink tubing, and resoldered them in order to make a safe connection.

With these jobs done, we re-tested the key switch and then threw the breaker on the Type 708 supply to the on position and turned on the PDP-8 for just one second, while testing the +10V supply rail on the backplane.

This was the first time the PDP-8 has seen power since around 1980.

We repeated this test, another one-second power-up, this time testing the -15V supply rail, and then doing the same thing to check the two memory supplies (R/W and Inhibit). During each of these tests, several lights on the front panel came on, further indication that something is working.

All the voltages are within the expected range, and the main +10 and -15 supplies are solidly at their indicated voltages. Therefore, we are ready to launch a new project: Testing the PDP-8 memory and CPU, Bug 43.

Feb. 5, 2015, Backplane switches

  thumbnail image     thumbnail image  
Memory backplaneSwitch wiring
  thumbnail image     thumbnail image  
Marginal check switches
Bug 43: The briefest test of the logic demonstrated that power is not reaching all of the backplane. Checking with an ohm meter demonstrated that the problem is in the switches along the bottom of the backplane. These switches select, for each power bus serving one row of the backplane, whether that bus is connected to the main supply or to the marginal check supply (which must, of course, be configured to put out the same voltage). Many of these slide-switches have contact resistances on the order of an ohm, and some do not seem to make contact at all. This leads us to declare a new bug:

Bug 44: We must repair or replace several of the marginal check switches at the bottom of the two backplanes. David Gesswein has documented the same problem with his table-top PDP-8, serial number 184. He solved this by disassembling the switches, polishing the contacts, and reassembling them.

Whether we repair or replace the switches, we will have to get at them, so we unscrewed the switch assembly from the bottom of the memory half of the backplane and then photographed the wiring before disconnecting it. In disconnecting the wires, we found that one of the Faston connectors (orange, +10V Marginal Check) was not properly seated. (You can actually see this wiring error in the upper right photo if you look closely at the fifth connector from the right end.)

Tilting the bar to expose the backs of the switches, we photographed them. Notice that the switches for backplane row C (the third pair of switches from the right, in the lower right photo) are not originals. These must have been replaced before the machine was retired.

The switches have the following specifications:

A quick check of eBay and other sources suggests that this is a $5 part, new, but possibly $0.35 from several surplus outlets.

Feb. 6, 2015, Backplane switches

Bug 44: More careful examination of the two replacement switches on the memory half backplane show that they are SPDT switches with provisions for manufacture as DPDT switches. The replacements appear to have no nomenclature or specifications embossed on their bodies, only the manufacturer's name Switchcraft. These switches appear indistinguishable from the Switchcraft 46203LSRX except that the handles are white. Searching eBay for Switchcraft slide switches finds numerous dealers who stock the DPDT equivalent with a white handle, marketed as a replacement part for several brands of electric guitar, all priced at about $5.00.
  thumbnail image     thumbnail image  
Switch backs, before and after
  thumbnail image     thumbnail image  
Switch frontsSwitch Buttons

Given the delay and price involved in ordering replacement parts, we opted David Gesswein's approach and rebuild the switches we have. We unscrewed all the switches from their frame so that the switches were hanging by their wires, and then pried open the ears on the switch that was the worst. This allowed removal of the front half of the switch from the contact assembly. The front half consists of a metal frame, plastic slider, spring, and contact button.

The button in the first switch we removed was mostly black, except for parts that were green. Both the black and green came off with isopropyl alcohol, revealing the silver-plated contact button and contacts. We polished the button (a few brisk strokes on tightly stretched denim were sufficient to bring it to a high shine). Comparing the result with the photos on David Gesswein's web page, it appears that we are lucky in that there is no visible erosion through the silver plate. After adding a dab of new high-grade grease to the contacts, we reassembled the switch and tested it. It was mechanically snappy -- it had had a gummy feel to its operation, and the contact resistance that had been near infinite was brought down to near zero.

Investigating other switches in the row, starting with those that seemed unreliable or had high contact resistances, revealed that essentially all of them were in bad shape. While the green is evidence of some corrosion, the major problem appears to be contact grease that has hardened to a consistency comparable to varnish. Doug Ives only left after he had rebuilt the entire row of switches, excepting those that had been replaced. His photos shown here illustrate the conditions of the switches before and after cleaning; the photo of the switch buttons shows several that are partially cleaned as well as one in its original state and one that is brightly polished.

Feb. 9, 2015, Backplane switches

  thumbnail image  
Switch work
Bug 44: We reassembled the switch-bar on the memory half backplane. Each switch has its center terminal bridged to ground through a capacitor. These capacitors should suppress any sparking if the switch is thrown while the CPU is under power, and they might even permit uninterrupted operation when the switch is thrown, so long as the marginal check voltage is close to the operating voltage. One of these capacitors was loose from the ground wire, either pulled loose during our switch rebuilding or never properly assembled. Doug Ives (on the left in the photo) resoldered it before we attached the switch bar back to the memory-side chassis.

Meanwhile, Zimu Zhang (on the right in the photo) set to work disassembling the switch bar on the CPU side so that he and Doug Ives could rebuild those switches.

Feb. 10, 2015, Backplane switches and fans

Bug 44: I tested the results of the switch rebuilding, reattached all of the switches, and then found one broken wire on the +10V switch for row E on the CPU side. After re-soldering that wire, I tested the electrical continuity through all of the switches from the faston connectors on the switch bar all the way to the appropriate backpane busses. Everything worked, so I reassembled the switch bar and plugged all of the wires from the power supply back to the matching faston connectors.

  thumbnail image  
The fan bar
  thumbnail image  
Fan hub opened
Bug 45: In the process of ducking under the backplane segments to reattach the switch bars, I noticed that the bottom sides of the fans were, in some cases, disgustingly dusty. After close inspection of a fan, I found that the blades are screwed onto the fan hub from the top, and that removal of the blades should allow a reasonable degree of access to the rest of the fan for cleaning.

To gain access to the top sides of the memory-side fan shelf, I removed the bottom row of cards from the memory-side backplane, after carefully recording which slots they had occupied and making an effort to keep the cards strictly in order so each card would go back in the slot from which it came.

With that done, I disassembled a fan. There are 4 screws in the top side of each hub. Two of them hold the fan blades to the hub, while the other two hold the "hub cap" to the hub. Do not open the hub until you have cleaned all of the dust. When I removed a hub cap, I was able to add a few drops of new oil to the hub bearings. I used Starrett 1620 Tool and Instrument Oil that we had purchased for Teletype maintenance. The photo of the open hub shows the hub cap sitting upside down on on the front right side of the fan shelf.

There is one tricky detail that opening the fan hub exposes. In operation, at least while at rest, the weight of the fan rotor rests on the end of the vertical fan shaft. The thrust bearing there appears to consist of a disk of polyethylene (or some other relatively slick plastic) that sits between the polished inside surface of the rotor's hub cap and the rounded and polished end of the shaft. If, on disassembly, this disk stays with the shaft, as shown in the photo of an open fan hub, it must be carefully re-centered before the hub cap is replaced.

Feb. 11, 2015, Fans and front panel

Bug 45: I finished cleaning and oiling the fans on the memory side, reinstalled the boards that I'd pulled from row MF, and then moved to the processor side of the machine. There, I repeated the process, pulling the boards from row PF, removing the fan from each rotor, cleaning both the blades and the fan housing, and then opening the hub so I could oil it.

Bug 43: Doug Ives and I then powered up the machine and began working with the front panel. The switches are balky, but the load address switch worked for clearing the program counter. Repeated use of the examine or deposit switch increments the program counter, although some of the lightbulbs don't light, and the memory address register tracks the program counter (one step behind), with other lights not working. The deposit switch loads from the switch register into the memory buffer register, with yet other lights non working, and the continue and stop switches toggle the run light. When the machine is in the run state, the PC is obviously incrementing. Enough lights are burnt out that it is difficult to see what is going on, so it is clear that further diagnostic work will require replacing a number of front panel lightbulbs. To do this, we will have to disassemble the front panel.

  thumbnail image     thumbnail image  
Glass removedPanel down
  thumbnail image     thumbnail image  
Front panel glass, front and back
Bug 46: To remove the front panel you unscrew the screws holding the aluminum trim on one side of the computer, pull the glass sideways a small amount, and then tip the bottom outward. At this point, the glass comes free. The front panel is reverse-painted (rather, silk screen printed) glass, not plexiglass as in many later computers. As such, care is required to avoid damage to the glass, it is brittle and the reverse-painted surface is easy to scratch. The body of the panel is a heavy masonite or particle board sheet, with galvanized iron inserts to hold the mounting screws. 4 screws hold the masonite to the steel frame of the computer, while 4 more screws hold the circuit board to the masonite.

Having removed the front panel, we can attack a number of problems. The burnt-out lights were our immediate motivation for Bug 47. We can also address Bug 1, the broken switch-register handles, and while we are at it, some of the switches appear to be intemittent and may need cleaning in the same way we cleaned the backplane marginal-check switches.

Bug 47: Each bulb sits in a hole in the masonite sheet, with the wire leads of the bulb soldered to the circuit board. There are a total of 76 bulbs in our front panel, approximately half seem to be burnt out. The machine has no bulbs in the 6 positions reserved for the memory extension control; apparently, to save money, these bulbs were only installed if needed.

Vince Slyngstad maintains a web page that lists the bulbs for various models of the PDP-8 and peripherals. He lists the original bulbs as part number 12-01741, which he lists as equivalent to 1762F/CM1762. In addition, he lists a longer life variant that is otherwise the same, the CM2187. (CM numbers refer to the old Chicago Miniature Lamp Company, now a division of VCC, Visual Communication Company.) David Gesswein has also reported success with the CM2187. The basic lamp specifications are:

We found the CM2187 priced at $0.89 each, $14.36 for 25, or $49.50 for 100 from Digikey. On eBay, we found it priced at $3.99 for 10.

Feb. 16, 2015, Front panel switches and Teletype

  thumbnail image     thumbnail image  
Switch handle repairs
Bug 1: As noted on Oct. 24, 2013 the 6 switch handles on the left end of the PDP-8 front-panel switch register are loose. We pulled these handles and found that, on each one (and in one case both) plastic lugs that serve as hinge pins had broken off. This is a well-known problem with the handles on PDP-8 switches, both on the classic -8 and on newer models built into the 1970s. David Gesswein has a web page that describes the same problem and shows his solution. Vince Slyngstad has a web page on his efforts to manufacture new handles. This page explains that Slyngstad was making handles for a PDP-8/E and describes the difficult he had matching the color of the originals. Aside from the color, the handles he made are mechanically identical to the handles for the classic -8, as can be seen by comparing his photos with those here and on Gesswein's page.

We opted to repair the handles using the method documented by David Gesswein. I had previously repaired PDP-8/F handles using the same approach. I drilled out the handles in line with the broken hinge pin and then inserted a length of 1/16 inch diameter brass rod to replace the broken hinge pins. The left photo above shows the handles after drilling, some with brass rods in place, while the white switch handles closest to the camera include two that are newly installed after repair.

Bug 26: We found a loose spring inside the Teletype. Where did it come from?

Bug 48: Also, we plugged in the Teletype for the first time since we got it. When the switch is in the off position, the paper-tape reader advance magnet engages. When the switch is in the local position, the machine chatters away at 10 characters per second, constantly. This is wrong, but for it to chatter the way it doesn, there must be a considerable amount of mechanism that is working correctly.

Feb. 17, 2015, Front panel lights and switches

Bug 1: We finished replacing the broken switch handles. We need to work on the intermittent switch contacts.

Bug 47: With the front panel open, we reverse engineered the lamp driver circuitry that is exposed. We have not found this front panel circuitry documented anywhere else:

schematic diagram

The components are:

There is one interesting note about the above circuit: Some machines that used incandescent bulbs in their front panels appear to have used "keep warm" circuitry to prolong the bulb life. On such a machine, the current in the bulb is never allowed to drop to zero; rather, each bulb sees either a high enough curent to light the filament, or a current that keeps the filament warm without lighting it. There is no evidence of such tricks here. When the logic input is pulled above the forward voltage drop of the D661 diode plus the emitter of the transistor, the transistor will turn hard off. This means that substitution of a LED with an integral current limiter may be feasible.

We tested one of the CM2187 bulbs we ordered, and it glowed nicely at 15 volts (we saw no reason to turn it up to the full rated 28 volts).

Feb. 23, 2015, Front panel lights.

Bug 47: Removing bad lamps is easiest with the circuit board pulled from the front panel, but pulling the circuit board must be done very carefully, as the lamps are very fragile. Once the bad lamps are removed, all remaining work is most easily completed with the circuit board mounted back to the masonite body. Re-attaching the circuit board is even more difficult than removing it because each lamp must be carefully aligned with its hole in the masonite body of the front panel. Failure to get a lamp into its hole can bend the lamp flat against the circuit board, breaking a lead, and it can even crush the lamp.

Removing bad lamps without pulling the circuit board requires use of a tool to pull on the lamp from the front while unsoldering it fromt he back. A simple wire hook suffices. After pulling the old lamp, a solder-sucker is just about the only way to clear the holes. This computer predates the use of plated through holes, so DEC placed eyelets in each hole. The openings in these gromets are small and long enough that solder wick has difficulty pulling the solder out of the center of the hole.

We finished replacing the lamps that tested bad, testing each lamp with an ohm-meter, replacing those that were open circuits, and then re-testing the lamp after replacement. One replacement lamp turned out to be bad and needed immediate replacement.

  thumbnail image  
Testing the Lamps
Building on the reverse engineering done on Feb. 17, we tested the lamps. To do this, we un-soldered the -15V lead from the terminal strip on the lower left of the back of the front panel, and unplugged all of the backplane connectors that linked the front panel to the power computer's backplane (7 connectors on the processor side, 1 connector on the memory side). This allowed us to power up the panel from a bench supply. Note, with the control inputs open, the bases of the transistors are pulled down, turning on the transistors. Therefore, applying power to the front panel in this configuration should turn on all of the lamps; a lit lamp indicates that both the lamp and its control transistor are both good.

Hooking the positve output of the supply to the black (ground) line, we turned up the voltage until the current hit 1.5A, full scale. This was 8V; at this voltage, there is just enough power going to each lamp to make it glow, as shown in the photo. The first time we did this, two lamps did not light. In one case, we had forgotten to solder one lead. Three other lamps were lit far brighter than their neighbors. We replaced the brightest of these with a new lamp. In total, we replaced 28 lamps.

The MA9 lamp (reporting bit 9 of the memory address register), the 3rd from the right in the second row in the photo, did not light. Using an ohm meter, the lamp tests good, and directly grounding the top connection on this lamp lights it. Therefore, the transistor that controls this lamp is not turning on. Further investigation of this will require removing the circuit board from the back of the front panel. Having done this once, we are not anxious to do it again, so we will leave this problem for later.

Note that one other lamp in the photo is not lit, the 5th from the top in the right column. This lamp is unused, as can be seen from the lack of a label in the front panel artwork photographed on Feb. 11.

Feb. 27, 2015, Public talk and demonstration.

In celebration of the 50th anniversary of the University of Iowa Deprtment of Computer Science, we had a dempartmental event, and the kickoff for this event was a public exhibition of the PDP-8. This involved moving the PDP-8 to a lecture hall two buildings away from the retrocomputing lab (across a snowy parking lot, through the engineering building, and across a street, with a total of 3 elevator trips and one wheelchair ramp. The hard-packed slush of a recent snowfall made parts of the move difficult.

In preparation for the event, I determined that we have spent, so far, $655 on parts and supplies for the PDP-8 and its peripherals. Some costs may have been overlooked, so $700 is a fair guess at an upper bound. This includes everything from replacement parts to cleaning supplies.

  thumbnail image     thumbnail image  
Lights onOne doesn't work
At the demonstration, there were two points when I was a bit surprised to hear a collecitve intake of breath from the audience. The first was when I turned on the power and the front panel lights came on. Alberto Segre took a photo of the machine with lights on.

Unfortunately, it doesn't execute instructions yet, but I was able to demonstrate that the program counter counts when I toggle the run switch, and that the memory address register tracks the program counter, one step behind. Mark Matel took the photo to the right showing me pointing to the memory address register light that doesn't work because of a burnt-out driver transistor.

  thumbnail image     thumbnail image  
The demoMoving the machine
The second gasp came when (after turning off the power), I pulled the machine out on its rack slides and opened the two halves of the computer to expose the wire-wrapped backplane. Hantao Zhang took a nice photo of the crowd gathered around the machine after it was pulled open.

Two alumni who attended the event helped move the machine back to the Retrocomputing lab, Charles Ryan and Mike Miller.