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.
Bug 4 and Bug 64: Given that quite a bit of the the CPU seems to be working, we began to investigate the memory. One of the first things we noticed was that the Maintenance Manual (Feb. 1966 Ed., pages 10-70 to 10-76) documents 7 W034 cables between the CPU and memory half backplanes, but there are only 6 present. The missing cable (page 10-76) would connect backplane slots PD02 and MD35. While we have not fully understood the signals that pass through the various cables, this makes it more pressing that we attempt to manufacture replacements for these cables.
|W034D cable connector|
Having done this, we found that Vince Slyngstad has already done much of this.
CAD Project Files,
indexed under ./DEC/Wxxx/W034: Flexprint, 16 connections on "B" side 10
ohms on A2, B2. This includes photos, schematics and board layouts for
the W034D, W034H, and W034X. The latter is a modern redesign, while the two
former are historical DEC designs. We need to evaluate his modern design to
see if it meets our requirements.
Bug 31: After poking around the ADC rack, we decided to remove the locally built magnetic tape drive interface from the ADC rack. Removal of this device interface is complicated by the fact that it is not daisy-chained into the I/O bus the way off-the-shelf device interfaces were done. Instead, wire-wrap wiring has been used to tap into the I/O bus connectors in the analog to digital converter backplane immediately above it. In addition, spare slots in that backplane have been used for a few additional logic boards and interface plugs needed for the tape drive.
|Using an unwrapping tool|
Pin E12B refers to a pin on row E (counting from the top), slot 12 (counting from the left side of the wiring side of the backplane) pin E (as printed in white on the wiring side of the black connector blocks). The tape-drive interface was mostly in rows E and F, with wiring to a few spare slots in row D and one slot (used for a cable connector) in row C. We logged the wire color, source pin and destination pin for each wire we removed.
We also pulled the cables that connected the tape-drive control panel to
slots in the backplane segments involved. The wires at one end of this
cable were soldered to parts of the control panel, another end of this
cable ended in cut-off wires that apparently went to the tape drive, while
the third and forth ends of this cable went to card-edge connectors plugged
into the backplane. At least one of these was almost certainly plugged into
the wrong backplane slot, but we wrote on each card-edge connector the slot
number into which it had been plugged.
|Removing rows E and F|
With the final wire removed, we were able to pull the remains of the home-brew tape-drive interface from the rack. We believe that some boards on the lower right end of row D are also part of this home-brew interface. Wiring to those boards from rows E and F has been removed, but the local wiring on row D remains in place.
The home-made tape drive control panel directly below rows E and F had one remaining cable still connecting it to other parts of the rack. In this case, to a block of relays attached to a board cantelevered out from the back of the topmost panel on the back of the rack. We simply cut the wires from the cable to this block of relays. The colors of the cut wires should be sufficient to allow reversing this change, should someone want to do so. We will tag all removed parts to allow this.
Bug 64: We tracked down David Gesswein's photos of a table-top PDP-8 he restored. These photos of his show the cabling between the two half backplanes:
These photos clearly show only 6 cables, all from Px01 to Mx36
(for x from A to F). The photos do not show the 7th cable connecting
PD02 to MD35.
On Feb. 12, 2018
we had speculated that perhaps this missing cable was the cause of our memory
problems. Apparently, this is not the case. It appears that this cable is
not needed on machines with only 4K of memory. Therefore, cable replacement
moves down on our list of prioritie, but remains there because of the decayed
nature of the 6 cables we have.
Looking at the detailed drawings in the Maintenance Manual, BS-D-8M-0-12 X Axis Selection, BS-D-8M-0-13 Y Axis Selection and BS-D-AM-0-15 Sense Amps, Inhibit Drivers, and Memory Control all show the detailed locations of the boards involved and document the role of 80&Ohm; current limiting resistors, but they don't show where the resistors are. We found these resistors lurking on an aluminum heat sink behind the balun network boards on the front-side of the core stack. See photo. These are all 80&Ohm; 10W 1% precision resistors. We infer from the fact that two of them have small resistor capacitor networks serving as shunts that those two are the X and Y axis current limiters attached to the G209 modules, while the other 13 resistors each serve one of the G208 inhibit drivers (the 13th resistor is a spare on our system since we don't have the memory parity option.) Note that this little RC network isn't documented in the maintenance manual.
A second problem we will face is that the above-cited manual pages all refer
to handwritten notes that should be on the label of our memory module. We have
found no such notes. The
has a photo of the label on
memory that gives a R/W (select) current of 325mA and an inhibit current of
305mA, so if we cannot find documentation for our machine, we at least know
reasonable values for these currents.
|Instrumented current limiters|
An alternative way to measure the current through any of the 80Ω
resistors is to simply measure the voltage across the resistor, but they are
inaccessible behind the circuit boards holding the memory
To allow this method of measuring the current, we attached a set of 4 test
points at an accessible location directly behind the plate on which the
resistors are mounted and wired a pair of test points across one of the
resistors limiting the select current and one limiting the inhibit current.
|Inhibit and R/W select currents|
The good news is that we got waveforms on both the R/W select and inhibit currents. The waveforms start when you hit the CONT (continue) switch on the front panel, and they stop when you hit HALT. This means that the key memory control signals from the CPU are reaching the memory. This is the first time we have confirmed this.
Checking the scope by using its internal calibration oscillator, we found
that it was way out of calibration, so we spent the rest of the afternoon
searching the scope manual for calibration instructions and then doing
what we could to calibrate it using its internal calibration oscillator and
a DC voltmeter. The scope was, in fact, way out of calibration. The internal
10V DC supply from which the scope derives all of its other voltages was under
8 volts. Adjusting this and then bringing everything else into balance
seems to have made the scope trigger circuitry work much better.
|Inhibit and R/W select currents
1µsec/div hoiz; 10v/div vertical
The Dalby Datormuseum photo of their their PDP-8 memory label that gives a R/W select current of 325mA and an inhibit current of 305mA, so the values we see are in the right ballpark.
Because this is core memory, each memory cycle consists of a read/erase subcycle where the contents of the addressed word are destructively read, followed by a write/restore subcycle where the data just read is returned. The inhibit line is only pulsed during a write/restore cycle, and then, only when the value written is zero. Since we are currently reading all zeros from memory on every cycle, the inhibit line is pulsed once per cycle. This looks right.
The R/W select lines are more complex. During the read/erase subcycle, the read select line should be pulsed, and during the write/restore subcycle, the write select line should be pulsed. These pulses are mixed by the balun networks on the G603 boards attached directly to the core memory module to produce a negative current pulse for the read/erase subcycle and a positive pulse for the write/restore subcycle. Both the read and write select lines in each pair share a single current limiting resistor, as shown in the Maintenance Manual (1965 ed, Figure 4-4, page 4-6). Thus, we should see two current pulses during each memory cycle, one for the read/erase subcycle, and one for the write/restore subcycle. We do see two pulses per subcycle, but they are strangely brief and they have different magnitudes! Something seems amiss.
|Inhibit logic signal|
|Read and Write logic signals
1µsec/div hoiz; 1.0v/div vertical
We used internal triggering for these measurements, so do not read any
meaning into the phase differences on the scope displays.
|Measuring supply ripple|
The photo shows our
as used in these measurements. As mentioned on
Dec. 16, 2014,
the scope is set up for differential measurement, sitting on a dry wooden plank
to isolate it from ground and running from an isolated DC supply. This is
required because the inhibit and R/W select power supplies are not regulated
relative to ground.
|Inhibit ripple at core and at supply
1µsec/div hoiz; 0.2v/div vertical
The high frequency component of the ripple measured at the supply output
is almost certainly an artifact of the feedback delay in the regulator circuit,
and it is noteworthy that this high frequency noise is invisible at the other
end of the several feet of
between the power supply and memory. What we do see at the other end is
a low-voltage echo of the supply current waveform we measured on
The relatively low resistance (or, at these frequencies,
of the twisted pair translates the current waveform to the waveform we observed.
|R/W select ripple at core and at supply
1µsec/div hoiz; 0.2v/div vertical
As with the inhibit current, R/W select current ripple at the memory is higher aplitude and reflects the current waveform measured on Apr. 18. Unlike the case with the Inhibit current, however, we do not see the high frequency artifacts of the regulator circuitry at the power supply output, but just the same miniature version of the current waveform. This leads us to suspect that something in the R/W select supply is not working correctly.
Note that, when we reformed the power supply capacitors
(Bug 10), we found, on
Mar. 13, 2014,
that C15 was bad, and on
Mar. 27, 2014,
we installed a replacement.
C15 is the final (post regulator) filter capacitor for the inhibit supply,
while its twin, C16, is the final filter capacitor for the R/W select supply.
Both are nominally 980µF 75V electrolytic capacitors, although our
replacement is labeled 1000µF.
If C16 capacitor is bad, it would lead to poor power supply regulation,
but we also need to consider the possibility that the series regulator
transistors are bad or that one of the G808 boards that control them is bad.
Bug 64: To help rule out problems with the power supply regulation, we tried switching the power transistors Q1 and Q2 inside the power supply. These are big 2N3715 80V, 150W, 10A Silicon NPN transistors mounted on the power supply frame. Q1 regulates the Inhibit supply, Q2 regulates the R/W Select supply. See the power supply schematic in the Maintenance Manual (1965 ed, Drawing RS-C-708, page 10-8). Switching the transistors had no obvious effect on the ripple signal observed across the power supply output, although in switching the transistors, we did notice that one of the mounting screws for one transistor was not very tight. This could have added resistance to the current path.
We also tried switching the G808 regulator boards in the power supply. Given that the set point of each board is adjusted differently, we would not expect the machine to work with these boards switched, but we wanted to see if the quality of voltage regulation changed. It did not, so we switched the boards back.
Conclusion. The power supply electronics is not obviously bad.