PDP-8 Restoration

[SierraFoxtrot]

For me, one of the real jewels of the ACMS collection is the original model PDP-8 (aka the straight-8). There is so much to love from the whopping 4k words of core memory to the wood grain paneling she is a thing of beauty.

At the end of September, @daveeverett and I took the perspex side panels off the cabinet and started a full inspection. Apart from dust and grime, everything looked to be in pretty good shape. So we set to work restoring it to working order with the first objective being the model 708A power supply.

Goal #0 for this project: To restore the PDP-8 in as original condition as possible choosing repair over replacement every time.

The power supply is tucked behind the glorious front panel and is somewhat accessible when you extend the computer itself out on the slides. Peering down in the supply, you get impression DEC got a pretty good deal on massive electrolytic capacitors back in the day.

Looking down from the top, you can see a few highlights:

1. The ferroresonant transformer and it’s matched capacitor (Blue)
2. Some of the 7 high current diodes bolted through the chassis (Gold)
3. The “power OK” relay to prevent erratic supplies corrupting core memory (Red)
4. The memory power supply regulators (Green)

SLOWLY SLOWLY…
The 708A features 15 silver coloured big boy electrolytic capacitors which have been sitting powered down for (likely) decades. If we were to throw 240VAC across the input at this stage, they would fail where the aluminium oxide has become thin over the years as it was eaten away by the electrolyte. The likely result being several large bangs. That means it’s time to reform them.

The basic procedure is to isolate each capacitor. That bit is made easier by the spade lugs! Then start by applying 10% of the working voltage. It’s extremely important to limit the to something small e.g. 10-20mA. I did this with a power supply that supports a controllable max current but a series resistor will also do the trick.

The setup for two at a time:

Assuming we don’t already have a dead short-circuit… the capacitor will SLOWLY charge up
to the supply voltage. If you look with a digital ammeter it will appear that current has ceased flowing. However a decent analogue ammeter like the AVO pictured will show that a small amount of current is still being drawn. This is actively reforming the oxide layer in the capacitor. After some time (minutes or even an hour), the current will flatten out and then eventually decrease towards zero. Bump the supply up another 10%, leave the current limited, rinse and repeat until you get up to the full working voltage.

What I’ve discovered: If you think you are rushing, you are. If you have to ask “how long”, the answer is longer.

A short video of the process

After a large number of Saturdays, we now have 60 year old caps operating pretty close to spec. Zero bangs, all original parts!

More to come…

[SierraFoxtrot]

I’m playing catch-up on journalling this so here is another entry to take us up to mid December 2025

Part 2: Turning it on…

With the caps refreshed, and hopefully no longer threatening to act as a dead-short, it was time to check out the rest of the supply and work towards powering it up.

Diodes

Most of the diodes could be tested either in situ or by pulling a spade lug to isolate them. Some however, required pulling out the socket set to remove them from the chassis. Nothing says power electronics like pulling out the real tools.

Impressively, these 1n3208 diodes are still manufactured and sold through DigiKey and the like. Given the maximum of 2A from any of the channels, their rated max of 50A does seem like overkill.

Regardless, testing ahead of time was going to avoid unpleasantness down the road so we took the new high “Multifunctional Tester” out for a spin. Thank you Michael for the crash course!

NOTE: While the basic diode setting on your Fluke (or Fluke knock-off) will check forward and backward bias with a Volt or two, this unassuming little gadget will generate a few hundred volts and let you know the reverse breakdown voltage of the diode under test. This is very handy for aging silicon which may be under-performing in terms of specification but will fly under the radar on a typical DMM.

I was almost disappointed when they all held off a couple of hundred volts. Almost.

At this point, we are running with all original diodes now too!

Transformer

A brief diversion i.e. a furphy led to pulling out the transformer to check for shorts. It was either that or keep tripping the main breaker for the whole building a few more times (sorry Sean & Tim!).

The transformer turned out to be fine, as was its companion capacitor. But doing a thorough inspection on these was definitely worth the time.

Pro-tip: Unlike the electrolytics we reformed above, these oil-filled capacitors of the era are chock full of Polychlorinated Biphenyl, the nasty kind of PCB. PPE is recommended!

Fortunately, both checked out so back in it went.

Firing it up

At the (correct) insistence of @DigitalRampage we just flicked the switch which rewarded us with the prettiest sight of all:

With the gremlins banished, the usual tripping breaker was replaced with the buzz of the transformer and the warm glow of neon. Absolutely delightful. The +10V and -15V channels were rock solid with a 1A load.

Given an allowable ripple of 700mV and that we can expect a bit of a voltage drop over the massive cable loom up to the computer, 40mV or ripple at 10.4V will do quite nicely for now.

This is fortunate as the maintenance manual states that if it is above 700mV of ripple, it’s faulty so just fix it.

Just before we go celebrating too soon, a quick check of the margin check voltages showed them way off spec and not responding to adjustments. But that had to carry over to a following Saturday just like a Star Trek cliffhanger back in the day.

[SierraFoxtrot]

Part 3 Fixing the missing channels

Margin Check Supply

In the PDP 8 power supply, the margin check channels exist to deliberately run the logic from an adjustable substitute supply so you can “margin” the machine and expose borderline faults. Basically, running with the voltages a little high or low, stress tests borderline components ideally to catch faults during periodic maintenance rather than normally running. A nice idea from a time when components were less reliable, computers failed as a regular event and hardware maintenance was something you planned.

In practice, you select the margin check supply in place of (either) the main +10 V rail or the main −15 V rail via switches on the computer’s backplane. Next you vary the voltage output up and down while watching for errors, crashes, or specific symptoms. A switch, dial and convenient meter on the front allows you to adjust and measure.

If the system only fails when you push the supply slightly high or slightly low, you have found a marginal condition (weak gain, leakage, poor connections, drifting reference points, etc). This is useful both for preventative maintenance (confidence that the system still has noise and tolerance headroom) and for troubleshooting intermittent failures that do not reproduce at nominal voltages.

Fixing our margin supplies

Last time, we observed that the margin supplies were both very low and not responding to adjustments. A quick look at the schematic (the relevant section is in green) shows some likely suspects.

Firstly, the variable transformer on the front (Blue on schematic) was patchy and indeed open circuit in many spots after decades of sitting idle. Fortunately with some liberal use of contact cleaner (we really should hit up WD-40 or 3M for sponsorship…) and repeated turning, it came good giving a nice smooth adjustable DC voltage on the front panel meter.

After premature self-congratulations (are there any other kind?), I spotted that the output on the terminals was still shoddy. That left literally one component, switch S1 (Red on schematic).

Switch Surgery

After contact cleaner failed, I was tempted to pop down to Jaycar and fetch a simple toggle. But remember that we are aiming for repair over replace. Figuring that I couldn’t make things worse, the switch was disassembled.

What happens to grease over 60 years? Apparently it turns to something resembling wax, coats everything and helps to oxidise all the terminals. So after some cleaning and buffing of the contact points, it all went back together without too much fuss beeping appropriately when tested with a DMM.

A further quick check with the multi-meter confirmed that the margin check channels were now switchable, adjustable and present at the output terminals.

SUCCESS!

A trial separation…

One thing I failed to mention in the previous update was that working on the supply had become nightmarish. Extending the computer out the front of the cabinet to get access to the underlying supply threatened to topple the whole thing over!

Because of poor cable management and a rather difficult mounting arrangement, removing the supply from the cabinet threatened to be irreversible. Interestingly however, the computer is fairly self contained. So with Murray and @DigitalRampage’s help, the computer and supply were separated from each other for the first time this century.

That brings us up to date. Work has now started on the power control panel responsible for routing AC power throughout the cabinet.

To be continued…

[mdtapp]

So cool, wood paneling! Hope you live-stream the first power-up !

[SierraFoxtrot]

Thanks Mike! The wood paneling is the most cooperative subsystem so far. First power-up will definitely be documented. Live-streaming may depend on how brave I’m feeling that day, but it’s certainly tempting.

[mdtapp]

@SierraFoxtrot

at

Any computer with an Hour-Run counter and built-in Voltmeter has my automatic respect.
I have a modest-by-comparison Cromemco Z2D restoration project in the wings for 2026 and will be leaning on your Cap reforming and transformer testing hints there for starters. Nice work documenting that.

[SierraFoxtrot]

I’ve only ever seen the exterior of those but I’ve read that they also have a nice linear supply with big ol’ electrolytics. What those supplies lack in efficiency, they more than make up for in charm! I’d love to read a write-up if you get the chance!

[m_thompson]

The RICM had lots of problems with the margin switches on the backplanes. After cycling them many times they worked OK. There are two versions of the schematics, one for the first 50 shipped, and another for the production units. Make sure that you use the right schematics. There are a few people who have experience with this machine if you need help.

[SierraFoxtrot]

That’s incredibly helpful, thanks @m_thompson! Sounds like hard-earned knowledge and I suspect you just saved me weeks of frustration. I’ll check out the switches and the serial number label for clues.

[m_thompson]

You can read the restoration blog for the RICM’s Classic PDP-8 here. It might give you some ideas for how to debug and restore your system.

DEC PDP-8 2024 Restoration

[SierraFoxtrot]

Sensational! I love that the DF32 is getting tackled as well which will definitely be a future challenge for us too.

[SierraFoxtrot]

Part 4 An on/off switch. How hard can it be?

Once you throw in a few storage devices and the like, DEC systems can span more than one cabinet and pull down more than their fair share of Amps. One particular joy is turning all of that on with a single key just like starting a car. With our PDP-8, that is done with the 834B Power Control panel.

Features from left to right:

1. The rear of a 100nF filter capacitor (these will annoy us later in the story)
2. Power source selection to choose between local, remote and off
3. The mercury contactor
4. Transient suppressors (not capacitors, although they could likely play them on TV)
5. A hefty circuit breaker
6. A nice red neon
7. Another 100nF filter capacitor (because bad things come in threes. Yes three).

Similar in function to a relay, this panel sports a gorgeous mercury contactor which uses an electromagnet to move a metal slug allowing a puddle of mercury to close the circuit. Check out the video to see it in action.

Video of the mercury contactor in action

While a relay would suffice, the mercury contactor switches much higher loads with ease, the contacts never arc (and so never oxidise) and it adds an element of drama to the whole affair with the imminent threat of broken glass and toxic liquid metal spills.

I was told there would be capacitors misbehaving…

A quick test of the filter caps showed that they seemed decent sitting around 100nF at the ripe old age of 60. No need to fiddle with these. Move along. Nothing to see here.

The power selection switch much like the one in our last chapter was not switching but the same tear-down rebuild as last time gave fantastic results.

The transient suppressors are mostly there for moral support by 2026. They look pretty so they’ll be staying and I’ll back them up with a a decent power strip with modern transient protection.

The circuit breaker does what it says on the tin and the neon glows warmly.

It’s a simple circuit so I turn on the breaker for a quick test before moving on and blip the lights go dark. I mean the building lights. All of them. I’ve trip the power for the entire building.

What has this got to do with capacitors?!?!

@DigitalRampage is transferring thousands of scanned pages across the network for archiving of manuals. Sean is configuring a rather complex Mac network on an old SE without even a backup battery. Tim is reading a 9-track tape which is probably de-laminating as it goes over the read head. And I trip the main building breaker. I believe tradition dictates that I should shout the bar.

Specifically, what I tripped was the Earth Leakage Circuit Breaker (ELCB). In Australia these are adorably called a “safety switch” and often there is one per building.

Remember that pair of 100nF capacitors? Well, stick them in parallel as this circuit does and it’s 200nF. Also, the 708A power supply has a third giving about 300 nF across the mains, which produces a relatively large capacitive current at 240 V 50 Hz. As these paper oil capacitors age they absorb moisture and develop real resistive leakage, not just reactive current. Modern ELCBs detect this leakage or any resulting imbalance between active and neutral and trip immediately. Not a problem back in '65 but in 2026 it’s a headache.

But, how do we fix this while keeping to our goals of repair & restore over replace?

These capacitors are a metal can which is soldered shut. I figured I could open the can, remove the innards and place a modern 33nF Y2 filter capacitor inside; close the can and no one would be any the wiser. How the heck am I going to open this thing that is basically a heatsink (the natural enemy of the soldering iron)?

I figured I’d ask my Dad, Ray Finneran who taught electronics for about 40 years (so, this was not his first rodeo). He is also the proud owner of a scope soldering iron. This beast dumps 30A through the tip and Dad made short work of it. We were pretty pleased with how quickly the solder melted. We then had a small panic as oil started to leak out. As previously discussed, oil filled capacitors of the era are full of Polychlorinated Biphenyl and not to be messed with!

Plan B

So, plan B was to leave those capacitors in the cabinet but out of circuit. A couple of terminal strips stand in for the contacts on the back of the caps. Meanwhile a couple of modern 33nF caps will keep a lid on the electrical noise. This is not period correct but it balances safety and functionality. It is also importantly, completely reversible should we come across a better solution.

This time for sure!

I asked politely and very loudly if anyone was in the middle of anything important. File transfers were allowed to complete and work was saved. Throwing the breaker was blissfully, delightfully uneventful. It just turned on. Like power supplies do. Repeated power cycling was satisfyingly dull.

We now have a power supply that is stable and can be readily controlled. Dare I say it, we are ready to move on to the main computer!

Let’s get the band back together

Which reminds me that the main computer was about 20ft away sulking in the corner having been ignored until now.

Thanks again to pit crew Murray, Tim and Adrian, the computer was finally reunited with the cabinet.

Next up, we move to the computer rack!

To be continued…