The TI 34010

From: Jgzabol_at_aol.com <(Jgzabol_at_aol.com)>
Date: Sat Mar 6 13:07:32 1999

In einer eMail vom 06.03.99 10:25:15 MEZ, schreibt Derek:

<<
> The TI 34020 is the (principally similar) successor to the 34010, and I
have
> designed in my company our basic bread-and-butter product of that time
> around that some 7 years ago. This is machine vision, i.e. digital image
> processing.
 
 I've read (I forget where -- maybe Micro Cornucopia) that the 34020 is
 basically the same as the 34010, except that it works with larger amounts of
 data at the same time. Were there any instruction-set changes?

     I really cannot tell, I do not recall the 34010 and do not hace any docs
about it,
    though I do have everything about 34020.
 
> I do not think it is advisable to emulate anything. If you want to build
the
> Perfect Computer (tm) I would suggest to use Linux as operating system, and
> write your own driver for whatever display hardware you see fit.
 
 :) Well, the number of games on the PC is a very seductive thing.

        Then, 340x0 will not help you at all.
 
 I think Linux is very good overall. I plan on installing it on my Macintosh
 as soon as the quarter (school) is over. But there are some other
 interesting OSs too, and many of them are even relevant to the list. I've
 heard good things about RSX-11M, RT-11, the various TRS-80 OSs, FLEX, OS-9,
 SK*DOS, and others. There was also a competitor to CP/M which has been
 praised (possibly by Allison) because it provided some important services
 that CP/M didn't. Unfortunately I forget if it was by Northstar, Cromenco,
 or Ohio Scientific.
 
 Admittedly, many of those are simpler than UNIX. But a few are real-time,
 which UNIX is not. Besides, almost any design is interesting in some way.
 
 I also have visions of putting together a Transputer-based system and
 somehow creating vast amounts of computer power out of thin air. :) From
 my reading, it seems there are two main drawbacks: 1) Generally, the
 Transputer depends on a host. 2) Parallel processors are very powerful but
 they won't solve every problem equally well, and they present some very
 subtle and nasty traps to the student or programmer. Maybe it's time for
 Tony to come back into this thread, since he seems to be the Transputer
 expert.

     I have done quite some work on parallel systems; the efficiency
     and ease of programming depends _TRMENDOUSLY_ on the
     problem you are looking at; some (like Monte-Carlo simulation) just fit
     perfectly well, others (like running canned software) not at all.
     Of course, it will be mandatory for high-performance computing in the
future,
     in fact is already today, to go parallel; in all cases I know at
tremendous
     work on the algorithm level parallelization.

     My personal preference is in shared-memory parallel processing;
     the transputers are message-passing parallelism, which I personally
     do not find very useful. But that depends of course on the problems
     and algorithms I am interested in.
 
> faster on the TI 340x0. The basic thing to know is that in those days
people
> were most concerned about putting graphics display lists on the screen,
> that is you have a list of triangles, lines, circles, and whatnot, and the
> hardware primitives (really microcode of course) on the TI 340x0 allow you
> to do that pretty efficiently.
 
 I understand display lists pretty well. Many vector-graphics machines used
 them; Sketchpad relied on them and Evans & Sutherland continued the trend.
 (I'd love to play with some of their machines!)
 
 Also the Atari 8-bit machines had a limited form, and you could say the
 Amiga has them as well. This kind of dipslay list relates to memory layout
 and interpretation, rather than defining objects in memory. But it still
 makes certain tasks absurdly easy, instead of very time-consuming.
 
> That in other word means, that _ON_ _THE_ _APPLICATION_ _LEVEL_
> you must be prepared to hand DISPLAY LISTS over to the graphcis processor;
> in those times, that was done by people like AutoCad, e.g.
 
 How does the 34010 do in handling bitmaps?

     The 34020 (and I believe the 34010 had that as well) deals with that
through
     special bit-blit operations, bit-block-transfer and simultaneous
logical/masking
     operations. This is much supported by VRAM hardware used at that time;
     the VRAM chips have bitblt features built in. The performance was exactly
     equal to the VRAM theoretical max throughput, i.e. the TI processor made
the
     best possible use of available memory technology.
 
 -- Derek
>>

      John.
Received on Sat Mar 06 1999 - 13:07:32 GMT