Duane Sand (duanebsand.delete@this.gmail.com) on July 24, 2022 11:39 pm wrote:
> Adrian (a.delete@this.acm.org) on July 24, 2022 10:07 pm wrote:
> > Adrian (a.delete@this.acm.org) on July 24, 2022 9:53 pm wrote:
> > > Duane Sand (duanebsand.delete@this.gmail.com) on July 24, 2022 11:50 am wrote:
> > > > A friend worked in cpu engineering at IBM in the early 80's.
> > > > He recalls that there was a very fast microcontroller
> > > > known as "the 5 bit engine" that ran its instructions several times faster than the contemporary 3033 and
> > > > 3081 370-range mainframes it was used with. It was used as the microprogram core of the desk-sized IBM
> > > > 3060 operator and maintenance console of the 3033. Its speed apparently caused as much of a stir among
> > > > IBM engineers as the 801 risc project. Google does not find much about this speed demon. On comp.arch
> > > > in 1999, Julian Thomas responded to Del Cecchi's comment about IBM microcontrollers with
> > > > "Then there was the infamous 5 bit engine used in the 303x consoles!"
> > > >
> > > > My friend recalls this engine as having a 7nsec instruction time. And that engineering teams
> > > > resisted using its approach to build fast microcoded 370's instead of lots of hardware.
> > > >
> > > > What was this amazing thing? How was it so fast? How was it used? Did it need water cooling,
> > > > also? What were the reasons to not use similar techniques for the mainframes themselves? Could
> > > > a similar design be used in our latest cmos chips, with similarly fast single-thread results?
> > >
> > >
> > > I believe that the "3060" name for that console might be wrong, as googling for
> > > it finds a much older IBM 3060 console, which was used in 1970 for IBM 370.
> > >
> > > At bitsavers, there are a couple of manuals from 1979 for IBM 3033, which
> > > describe the console "IBM 3036", which appears to be that meant by you.
> > >
> > > Unfortunately, while those manuals give a detailed description of the external behavior of the IBM
> > > 3036 console, they do not give any information about how its microcoded processor is implemented.
> > >
> > I want to add that, if the clock frequency was 140 MHz, then there is no doubt that the processor
> > was implemented with MSI (Medium-Scale Integrated) oxide-insulated ECL (Emitter-Coupled Logic)
> > circuits, like the Fairchild 100k family (launched in 1973) or the other ECL families equivalent
> > with it, which were introduced later by Motorola and a few others.
>
> Thanks, yes I did mean to write 3036, the console for 3033.
>
>
>


The main IBM 3033 system was made mostly with logic integrated circuits manufactured by IBM itself, which seem to have been denser, i.e. with more gates per package, but slower than the commercial off-the-shelf ECL integrated circuits.

With the OTS MSI ECL ICs available at the time when IBM 3036 was designed, it would not have been too difficult to design a microcoded processor based on a 5-bit RALU (registers + ALU), with a clock frequency of 140 MHz, and which could have been implemented on a single large board.

The high clock frequency would have been easily achievable when the microinstructions had only 5-bit wide operands.


I would guess that indeed the CPU of the 3036 console was made with commercial OTS ICs, and this is the reason why any suggestions to apply a similar technology to the main CPU would have been rejected, because at that time it was a policy of IBM to make products using mostly ICs from the internal production.