I find it somewhat interesting that this an octal machine, but the word-size is 16-bit (not a multiple of 3).
The program-counter is 15-bit (32K address space), while the other registers are 16-bit. It seems, there is a “universal” indirect addressing bit (like the PDP-1 had it) and also a unversal indexing bit.
Here are the word formats as from the brochure:
This always seems strange to me, as well, yet both the PDP-11 and the DG Nova are 16-bit octal machines. I am not as familiar with the Nova as the PDP-11, but in the case of the -11, the instruction set is actually strongly octal (the register and mode fields are 3 bits each and fall cleanly on octal boundaries). Integers and addresses certainly have an annoying max of 177777, although the UNIBUS does have an 18-bit hardware address space.
I always wonder whether this is a symptom of the designers having worked on so many 18- or 36-bit computers prior.
Here it gets even weirder: with the addition of the index and indirect bits, I would have expected 17 bits for addressing, but it’s 16-bit.
(Regarding the designers, mind that bits are numbered from left to right, MSB to LSB, just like DEC did it in the early days. I wonder, if there’s some heritage involved.)
The 16-bit Norsk Data computers (founded in 1967, the first model came out in 1968) were also octal in the sense that the instruction set and just about anything else with numbers was octal. I can still read the most common octal instructions, from when I used to patch in new code in drivers in the past (up to the early nineties). Octal doesn’t really fit well into 16-bit words, as has been said, but the 16-bit instructions often had 3-bit subfields and then octal works well.
The Nova was basically the 16bit PDP/8 sequel that Olsen vetoed and the staff walked to found DG.
A lot of machines had octal leaning instruction sets - it’s often just a product of having 7 or 8 registers. The 8008 was programmed in octal. The 8080 isn’t but is actually much easier to work with in octal because the instruction decode is 2bit/3bit/3bit - eg 01xxxyyy is all the load reg,reg instructions (where (HL) is also encoded that way). 10xxxyyy is ALU with xxx the op and y the register and so on…
What was really demented was the 8008 address space. Because people worked in octal and the address was loaded in halves it was often the case addresses were written as things like 002:120
with the first octal number the number of the page (ie of 256 byte chunks). Very strange.
The canceled DEC PDP-X project was long rumored to be the origin of the Data General Nova since both were designed by the same people, but when the project’s design documents finally became available an evaluation showed that this was not the case.
