Re: FAA To Finally Ditch Floppy Disks & Win-95

On 6/17/25 12:00 AM, Rich wrote:

John Ames <commodorejohn@gmail.com> wrote:

On Sat, 14 Jun 2025 00:29:12 -0400
c186282 <c186282@nnada.net> wrote:
>

I've done TMS-9900 programming ... odd but interesting chip. Seems to
have evolved when there was little diff between on-chip cache memory
and main memory (not the tech to PUT a lot of ram into the CPU
either). Also had a sort of hardware solution to multi-user/multi-
processing which was very unique. Still remember "BLWP" - Branch And
Load Workspace Pointer". The 990 minis weren't so bad.

>
Always found that one intriguing. Yeah, like the 6502's zero-page, it's
a design from a different era as far as CPU-vs.-RAM-speed goes, but a
clever design-around for fairly elegant multi-tasking in light of it.
Oneathesedays I wanna take one of the later iterations (the TMS99105,
IIRC, is the last one that kept the memory-resident register-file
property; some later TI microcontrollers borrow the basic architecture,
but ditch that) and throw together a little homebrew hobbyist system...

 A similar effect played out in the HP3000 stack based CPU systems [1]
 The stack was kept in memory, and the CPU only had "registers" for the
top two, four or eight stack slots.
 But reality was that all the way up until somewhere around the iapx286
time range, RAM memory was faster than the CPU's it was attached to, so
large register files on the CPU, or giant caches, were not needed.  The
CPU was as fast as it was, and the RAM wasn't what was holding it back.
After sometime around the 286 time range, CPU speed started greatly
outpacing RAM speed and large on chip register files and caches (the
bigger the better) came into play.
 [1] https://en.wikipedia.org/wiki/HP3000#Use_of_stack_instead_of_registers
 

   The TMS-9900 was from an odd place in the technology
   where large on-chip cache was not possible - so it
   pushed most of that out onto ordinary system RAM.
   The positive in that was hardware-supported multi-user
   multi-processing became easier.
   I think the 9900 would support 256 users, and multiple
   processes for each user, all with hardware support.
   Today it's all done with software, but THEN ...
   Hmmmm ... what could we do with direct MU/MP stuff
   built into chips these days ?
   For the era, the TI-990 minis were pretty GOOD minis.
   Can't really complain. They were just, well, a bit
   ODD.