#nova 1200

I found the culprit behind the spurious /INTR signal from the Cassette I/O board that's been holding up the Nova 1200 Logic Test from completing successfully.

This 7474 is ancient, and has no pull-up resistor on the /PRESET input (which is fine for the original variants, though not recommended). The Q and /Q outputs (cyan and magenta) should be mirror images of one another, and they are not.

So I replaced it with a new 74LS74 part, which meant I had to add in that /PRESET pull-up resistor:

Testing the old 7474 confirms it's a piece of shit:

I ran the Nova 1200 Logic Test and... it passes. Repeatedly! Address 00121 contains the number of successful passes, which is over 8000!

Okay, let's go for broke: loading up Data General Single User BASIC...

HOLY SHIT IT WORKS!

On my Nova 1200, both the Cassette I/O board and the CPU board, were having issues passing serial tests. The /DS5 signal had incorrect logic levels, so I swapped some gates there and solved that problem.

Then the big problem, which took alot of work to diagnose and fix is specific to the CPU board. Using the Nova 1200 Logic Test tape (as in a virtual paper tape), tests were performed but no clear answer was forthcoming. It was failing, but the serial output messages weren't working right to tell us what failed, or now. I single-stepped through the program until we hit a test that didn't pass: the auto-incrementing addresses.

On a Nova, addresses 00020 thru 00027 auto-increment values when accessed. Meanwhile, addresses 00030 thru 00037 auto-decrement values when accessed. These functions worked, but unfortunately they were being activated on addresses 00120 thru 00137 which is not right at all.

It took days of tests, writing our own testing routines, and lots of oscilloscope captures before it was finally found. The N8859A quad-input NAND gate that is part of the address decode logic had started ignoring the /MBO13 and /MBO14 signals, meaning it was decoding incorrectly. That part has been replaced with a 74LS20 (despite technically requiring a 7440, Fairchild 9009, or the aforementioned Signetics part, which has no datasheet) -- it's fine. So, that problem finally got solved, but it was a pain because it's only using a single 4-input NAND gate to decode a 16-bit value for a 15-bit address. DG pulled this off by cycling through one nibble of the instruction at a time, synced with the four Processor Time Groups (PTGs).

See those cyan and magenta traces? When those go low, they should trigger the blue trace at the bottom to go high. There it is.

Sometimes you don't have the part you need. Sometimes it's because the pinout is different from the closest equivalent you have on hand.

Turns out that a common 7402 quad NOR gate doesn't match the pinout of an old Signetics N8885A quad NOR gate: they switched around one of the input pins and the output pins of each gate.

Alright Nova 1200, your turn. Let's figure out why the /CONRQ signal isn't happy...

Except for some reason it works now. So I put the front panel cover back on, gave it a bit of a rest to make sure it wasn't a temperature issue, and tested again and it's still fine.

Next problem: test the serial functionality with two programs:

One that spits out ASCII barf [check] One that echos whatever was typed into the terminal [check]

Well, both pass, so now it's time for the bootstrap loader, followed by the binary loader. Those seem to work. Let's try 095-000036-01 Nova 1200 Logic Test from 1970 to 1974. They built it for exactly this purpose.

Passes the first check aaaaand... fails the second. Okay, let's work with that. Later, serial tests stopped working too. Great...

See that blue line spiking? That's the input that triggers things, and it's subtle analog bullshit, which is then turned into clean digital high/low. It's very unorthodox compared to how we do things today.

So this turned into a more complicated issue, or set of issues. The front panel started getting mad again (/CONRQ again). And after studying the circuitry for awhile, I deduced it was from the 5V rail being below the safe threshold for the logic to work. The 5V rail delivers at most 12A to this computer per the specs, which is for a few hundred logic IC's (lamps and core memory use different voltage rails). That means if it isn't adjusted right, the voltage sags pretty badly when you get far enough away from the PSU.

That's on the front panel. Not ideal. It's not much different on the CPU board. The problem is that if I boost the 5V adjustment trimpot on the PSU to reasonable levels, the -5V and +12V rails climb.

-5V isn't so much of an issue, not much needs that to work. 12V is used by only 2 things, neither of which is logic: the front panel lamps (which are rated for safe operation between 11-17V per Data General), and the industrial Noctua fans in the PSU chassis.

When 5V is happy, 12V is more like 13.5V, which is liable to degrade and eventually destroy those fans. So I built this little thing to protect those fans: a pair of diodes to drop the voltage down just below 12V. We don't need to worry about the lamps, they're fine

Fixed my Nova, broke my Nova, fixed it again, broke it again.