[sdiy] Re: Bottom Ten ICs

[Richard Wentk]

At 19:25 07/10/2004 +0200, Theo wrote:
>Hi Richard,
>Inline.
>
>----- Original Message -----
>From: Richard Wentk <richard at skydancer.com>
><snip>
> > And why, if the design was that smart, why did TI's engineers then cripple
> > it with an 8-bit external bus?
> >
>It was a 16 bit processor with a 16 bit bus, but could address only 64Kb
>like the 8 bitters did.

Okay. I see part of the confusion, which was the 99/4 machine TI built 
around it. That definitely did have some *eccentric* features, including an 
8-bit data bus. (Why?) Clearly it didn't show off the architectural genius 
of the 9900 to best advantage.

Anyway, I thought I'd check out timings, and nothing has changed my mind. 
Many move operations needed 4 memory cycles - painful when using DRAM 
waitstates - and some needed as many as 10. The 6809 and Z80 were both more 
highly optimised, and if they'd had 16 bit data busses too I'd be surprised 
if they hadn't run significantly faster. The 6809 also had a *much* more 
creative and useful collection of addressing modes.

The bottom line is it was a backward looking architecture. One reason you 
can still buy x80 and 68x derivatives today, while the 9900 is just a 
footnote,  is because the designers had the foresight, or perhaps the luck, 
to understand that micros needed an architecture that wasn't just an 
attempt at a sub-PDP11 cut-down mini.

The RAM thing must have seemed like a good idea at the time, but it can't 
have been that hard to foresee that access times were going to go down 
fast, and once they did the main (alleged) attraction was going to become a 
serious liability.

> > The whole thing seems like a classic case of wood for trees. One unusual
> > architectural feature makes no sense on its own. You need everything
>around
> > it to work too. All the way up to the s/ware support and marketing levels,
> > which is another area where TI crashed and burned.
> >
>If you say so...

I do. While it's fun to noodle around with components and solder, twe all 
know that if you're not just doing something as a hobby, how you sell the 
products of your innate design brilliance is a lot more important than how 
brilliant a design really is. And TI were amazingly bad at selling the 9900.

The 990 did sort of okay, the 99/4 did slightly better, then Commodore gave 
it a good kicking and IBM flattened it to the extent it disappeared from 
everyone's radar. The wisdom of selling a 'home' micro that was really just 
a toy for $1500 - multiply by ~5 to get today's equivalent - and not giving 
away cheap developer licenses was more than a little questionable.

>No The TMS9900 was for the big bulky multi-processor machines.
>There even are addressing modes to access the memory of and execute code
>from other TMS9900s.

There are?

>Considering the 64k of addressable control lines the main target would be
>industrial control applications.

Actually I'm pretty sure it was a 12bit control line space. But anyway.

I have no doubt it was entirely excellent at process control. I'm less 
convinced that it was what the world needed at the centre of a home micro. 
So far as anyone can tell it didn't take the world long to become 
unconvinced too. ;)

Richard