The following is a transcript of a recent Delphi TI Net conference. Enjoy it.
Please do not distribute it in part, or without the accompanying Delphi signup
information. Several others will follow if everyone abides by these simple
rules. Also, please do not reply or followup this message with lengthy quotes
from the conference transcript -- especially DO NOT REPLY with lengthy quotes.
I have the entire transcript at my disposal, so I don't need large pieces
thrown back at me. Thank you for being considerate.
Jeff White jhwh...@ulkyvx.bitnet
cut here-------------------------------------------------------------------
TI INTERNATIONAL USERS NETWORK
CONFERENCE TRANSCRIPT
Karl Guttag
of Texas Instruments
Designer of Video Display Processors
Microprocessors
Graphics Processors
and More.
Sunday, September 20, 1992
GROUP LIST: 21:33:07
39) Karl Guttag Conference
JeffW, Don O'Neil, Dan Eicher, KarlG, Joe S., JERRYC, Brad Snyder,
JeffG, HYPERION
- idle
AVAILABLE LIST: () = in conf
MARCLEVINE, (TELEDATA/W), (JHWHITE/CFHY), (MYARC/R), LEIGHTYM,
YPSILON,
C_BOBBITT, (JSYZDEK/HY), (DONEIL), (BRADSNYDER), (EICHER), (JERRYC/A),
(HYPERION/X), CAL47
------ [14 in this area]
.JeffW> Karl Guttag join Texas Instruments in 1977. He worked on the
designof the 9918(A) video display processor, was the IC
architect of the 9995 and 99000 microprocessors, and specified
the design of the 340 family of graphics processors. The
9918(A) is used in the 99/4A and other computers of the early
80's. The 99/4A-compatible MYARC Geneve uses the 9995.
Several upscale video platforms running Microsoft Windows and
X Window Systems use 340-based boards to speed graphics.
340-based video boards are available for mid-level systems as
well, such as the Commodore Amiga.
.JeffW> Karl Guttag is currently the strategy manager for TI's
Computer Video Products. We welcome him tonight to discuss
past and present projects he has worked on for TI. Karl, do
you have any opening words? Or are you ready to take
questions?
.KarlG> Nothing much other than I am new at this conference so please
bear with me. ga
.JeffW> Karl, could you run your story here in conference? ga
.JeffW> If your TE has an ASCII upload, you might use it. ga
.KarlG> OK, I will give it a try
.KarlG> From Sprites to the TMS34010 Graphics System Processor I lot
of people when they think of TI and graphics, think of the
9918 and sprites (the term "sprites", a greek fairy, was
coined by one of the 9918's definers). Some people have asked
about why the 34010 did not have sprites, and my answer is
that the 34010 was designed to handle many more "sprites" and
with more capability than the 9918 ever could, but it does
them differently. It so happens I was one of the designers of
the 9918 and I was in charge of the definition of the 34010
-- and yes there is a connection. Below is that story.
.KarlG> Perhaps the most outstanding feature of the TMS9918 and 9928
(the 9928 was digitally identical to the 9918 but it had
different analog outputs) VDP's developed in 1977-78 were the
sprites. The sprite concept meant that one could move an
object around the screen by merely changing the X and Y
coordinate. The '18 was designed in the days when the
microprocessor's processing power was very limited and thus
"bit-blting", the bit aligned movement of pixel arrays, was
not practical.
.KarlG> While sprites were great for games (the 9928/18 were used in
Colecovision, TI's home computer, the Japanese home computer
standard MS/X, plus a vast number of consumer video products),
they have their limitations. For every sprite to be displayed
on a given line there had to be a set of hardware which
included shift registers, color latches, and counters. The
9918/28 had sprite pre-processing logic (that I designed) that
sorted through a list of 32 sprites to show those sprites that
would appear on a given line AND kept track of priority (who
was on top of who), but due to hardware limitations only 4
sets of display hardware existed. Also limits on the amount
of hardware meant that a given sprite could only be 1 color
and the size of the sprites were limited to 16 pixels wide
(this could be magnified by 2). To have more colors per
sprite would mean more shift registers and to have wider
sprites would mean longer (bigger) shift registers.
.KarlG> On top of the hardware requirements of sprites, there were
bandwidth issues as well. Sorting through the list of sprites
consumed a number of the few available memory cycles. If the
sprites where bigger or had more colors there would not have
been enough bandwidth to support over the 9918's 8-bit bus.
Did you know why the 9918 never supported full color
bitmapping? - it was because there was not enough memory
bandwidth available.
.KarlG> After the 9918, I went off to architect the TMS9995 and
TMS99000 16-bit microprocessors for the next 3 years -- they
were well designed chips but were late and not well marketed
(oh well). In late 1981 I was considering what I wanted to do
next, when an application group from TI in England claimed
they could do "bit-mapped sprites" (generate sprite like
objects on a bitmapped display) IN SOFTWARE with a 9995. The
9995 has an 8-bit external bus and 16-bit internal bus with a
16-bit-wide 256-byte on-chip 333ns RAM which could be used to
store tight loops (which made it a lot faster than an 8088 at
some things).
.KarlG> Even so, I thought they must be wrong (heck, I knew what they
both could do), but I pulled out a calculator, pencil, and
paper to see. The calculations showed that the 9995 could
handled a few "bitmapped sprites". The reason was that while
the 9918 would have to construct each sprite, line by line,
which wasted lots of time and memory bandwidth, the 9995 could
be much more organized in moving "bitmapped sprites." -- for
example, the 9918 had to constantly re-fetch information on
each scan line to define a sprite, where the 9995 would work
on each sprite from start to finish into the bitmapped frame
buffer.
.KarlG> The nasty problem with the 9995 handling sprites was the bit
alignment and field extracting and inserting problems in the
moves, which severely restricted the number an size of sprites
it could handle. The 9995 could not do everything that the
9918 could, BUT it could do some things that we would not have
tried on the 9918 such as multicolored sprites and big
sprites.
.KarlG> I then started to consider what could happen if one designed a
processor that could handle the bit-aligned manipulations. I
worked out that you could generate such sprite effects that it
would be prohibitive to build a "hardware sprite chip" that
could even come close. You could have more sprites on a line
and each of them could be multi-color and not have the severe
size limits that hardware imposed.
.KarlG> It turn out that some of the operations "bit-mapped sprites"
required already had a name. Xerox had coined the term
BitBlts (bit block transfers). In the same time frame as I was
calculating "bitmapped sprites" and what they could do,
Xerox's Smalltalk graphics kernal article appeared in Byte
(August 1981 pp 168-194). I was hooked, and the rest, as
they say, is history.
.KarlG> We had very lofty goals for this new chip concept. We wanted
to go beyond what was already being done. Particularly there
was NOTHING we could find written about COLOR -- so we started
creating. The first problem to solve was "TRANSPARENCY", this
was the concept that while you might define a object by a
rectangle, there could be "holes" in the rectangle defined by
place-saver "transparent" pixels (we had "transparency" on the
9918 and we were not going to give it up -- but note a number
of the "new" graphics chips DON't support transparency). We
invented (or re-invented as some of these concepts may have
first been thought of elsewhere but had not been published
that we could find) arithmetic operators, plane masking,
binary to color expansion, and transparency. In the end, we
called the new set of color options (including the boolean
operations) PixBlts, since we thought of them as pixel rather
than simply bit operations.
.KarlG> When our people were writing the MS-Windows driver, they
needed a "hardware" cusor (essentially a "hardware sprite").
They used a technique I developed back when looking a
bitmapped sprites. What you do is just before the cursor is
to be displayed you save the image under where the cursor is
to be drawn, then inhibit drawing until the lines having the
cursor have been displayed, and then un-draw the cusor by
replacing the saved image. This technique may sound on the
surface wasteful (and it is, there are more efficient
methods), but on a 640 by 480 display with a 16 by 16 cusor it
only takes up about 3% of the total bandwidth (ie you could
but up many cursors this way). And the net effect is just
what you would get with a "hardware cursor" only you can
support more of them with less restrictions than any chip I
know of that has "hardware cursors" -- ah, but that is what
the 34010 was about in the first place.
.KarlG> --- the end -----
GROUP LIST: 21:45:03
39) Karl Guttag Conference
JeffW, Don O'Neil, Dan Eicher, KarlG, Joe S., JERRYC, Brad Snyder,
JeffG, Hype, CAL47, GENNA, Michael., JCARVER, C_BOBBITT, TIMTESCH
- idle
DUNNELLS
AVAILABLE LIST: () = in conf
(TIMTESCH), (TELEDATA/W), (JHWHITE/CFHY), (JCARVER), (MYARC/R),
(LEIGHTYM),
(GENNA), (C_BOBBITT), (DUNNELLS), (JSYZDEK/HY), (DONEIL),
(BRADSNYDER),
(EICHER), (JERRYC/A), (HYPERION/X), (CAL47)
------ [16 in this area]
.JeffW> Okay. Here is how we normally work questions and answers in
TI Net conferences.
.JeffW> If you have a question for Karl Guttag, our special guest
tonight, type "q" or "?" and the moderator, which is I, will
recognize when it is your turn. Such as "ga,".
.JeffW> The guest will answer your ?, end his response with "ga", and
I will tell the next person in the ?-queue to ask his/her
question.
.JeffW> Let's get to it.
.JeffW> ga, Hype. Ask your question. ga
.Hype> karl, glad to hear that 340 has bitblt/sprite orientation of
it functions as well as just line draw etc. Curious: Can it
be added/integrated with any graphics board type (ie VGA, XGA
as well as TIGA ) ?
.KarlG> Normally, a TIGA/340 board runs with a VGA-pass though option.
Generally, the old DOS aps run on the VGA, while things like
MS-WINDOWS runs on the 340. ga
.Don O'Neil> ?
.Hype> Follow up/clarification?
.JeffW> ga, Hype with your follow-up. But please keep it brief. ga
.Hype> I am just trying to get an idea of how 340 will fit into
future of highend graphics (couldn't think how to phrase,
thnx)
.Hype> ga
.Dan Eicher> ?
.KarlG> The 34010 was introduced in 1986 with the 34020 being
introduce around 1989. They are primarily used today as
MS-Windows accelerator X-windows Terminals. Also they are
used a lot for PC CAD.
.KarlG> I'm now working on a revolutionary new processor that is
target at image processing, which will be the next big future
market for the desktop. ga
C_BOBBITT> ?
JERRYC> follow-up Hype - then Chris ga
JERRYC> Ga Chris then Don and Dan
C_BOBBITT> Oh,thanks.
C_BOBBITT> Lots of questions Mr. Guttag, but will only ask one this
time <smile>.. I'm no engineer, but I was wondering how
difficult it would be to use.. a 34020 or some variation
as a stand-alone microprocessor (albeit with a..
graphical bent). Would make one heck of a Nintendo
machine :> - ga
.KarlG> Yes, definitely it can and could be. The 34020 is a full
fledge programmable microprocessor. I tried for 2 years to
sell it to several game manufacturers a few years back as a
super game/ simulator/ virtual reality (abeit simple) machine.
At the time they were afraid to take on Nintendo. ga
.Don O'Neil> With the architecture of the 34 series processors, do you
believe that it is possible to emulate a 9918 and/or 9938
using the 34010/20?
.KarlG> I guess you could, but it would not be very practical. There
might also be some screen hickups since the 340 might not be
able to do everything in real time. Another problem would be
some of the hardware interfaces to the host, by the time you
cleaned it up, it would be cheaper to just buy a 9118 (the
problems are somewhat like trying to emulate a 8086 with a
68000). ga
.JeffW> ga, Dan
.Dan Eicher> Karl, even to this day. The 99xx has one of the richest
instruction sets available. Very little changed from the
9990-9995, then with the addition of the 99105 and macro
store, users could in essance write their own micro code.
Had there been a 990000 what direction do you think it
would have taken?
.KarlG> Boy, a lot of attempts and millions of dollars where spent at
TI on that question. The first thing that has to be fixed is
the address reach, and the second would be to move the
"workspace registers" into a cache. Note, the SPARC from SUN
basically has a workspace cache. Other things like more
registers and at least 32-bit registers would have to be done.
ga
.Don O'Neil> ?
.JeffW> ga, Don
.Dan Eicher> ?
.Michael.> ?
.Don O'Neil> Why did TI "forget" about the 99000 series? When it was
first introduced, it was _the_ hotest performer on the
market.
.KarlG> Basically, it was too little, too late when the 99000 came
out. I had a 68000 under a microscope before I started on the
99000. Intel and Motorola already had big support programs
behind their chips, while at TI we had never generated much
software for the microprocessors (the systems division
developed minicomputers that ran the same basic instructions,
but had different memory management so that almost none of
their software would run on the 9900/99000). Yes the 99000
did burn rubber in its day; it could access memory every 166
ns (the 68000 could only cycle memory at 400ns). But this
fast bus speed was necessary to get any performance out of the
memory to memory architecture, but it then required us to use
fast memory. I have often lamented that if we could have had a
register to register architecture with the 99000's design we
would have had a real barn burner. ga
.Don O'Neil> ?
.JeffW> ga, Dan
.Dan Eicher> With home electronic becoming more and more complex...
Has TI ever thought about re-releaseing a 99xx compatable
chip to replace the TMS-1000 as an embeded controller,
using cheaper silicon manfacturing technology? Any
thoughts on the 68689 ASIC Version of a 9995?
.KarlG> No. TI did do a version of the 9900, called the 9940, in
1976-78 that was a bit of a disaster, that was a 16-bit
microcomputer. It was both too early and a very poorly
designed chip. They finally got it working with a complete
re-design but by then the market had past.... I don't think
there is much chance of reviving the 9900 family in the way
you suggest. Note TI is very involved in building 16-bit
microprocessors today. There have been some announcement, for
example, a major program called PRISM which is a 16 bit
microcomputer with a very modular architecture. (the first
applications for PRISM will be in automotive) ga
GROUP LIST: 22:25:54
39) Karl Guttag Conference
JeffW, Don O'Neil, Dan Eicher, KarlG, Joe S., JERRYC, JeffG,
HYPERION,
CAL47, Michael., JCARVER, C_BOBBITT, TIMTESCH, TIRUGED, MARKANDERMAN,
SHH
- idle
AVAILABLE LIST: () = in conf
(TIMTESCH), BEEE, (TELEDATA/W), (JHWHITE/CFHY), (JCARVER), (MYARC/R),
(LEIGHTYM), (TIRUGED), (MARKANDERMAN), (C_BOBBITT), (JSYZDEK/HY),
(DONEIL),
(SHH), (EICHER), GPEARSON, (JERRYC/A), (HYPERION/X), (CAL47)
------ [18 in this area]
.JeffW> ga, Michael
.Michael.> Is there any plans in the future for the 99XXX series of
processors, or has TI let that line die?
.KarlG> I understand that the Military Group still develops some 9900
compatible processors, but that is about it (the government
has very long design-in design-out cycles, that their
contractors support). ga
.Dan Eicher> ?
.JeffW> ga, Don
.Don O'Neil> Do you know of any plans to introduce an ASIC version of
the 99000, similar to the SM68689 (asic version of
TMS9989)? If not, is it possible to do such a thing
(providing funding is there)>
.KarlG> As far as I know, TI has no plans. Generally we would need to
see a market for a few 100K units to entertain such a plan.
TI is more focused on future products. ga
.JeffW> ga, Dan
.Michael.> ?
.JeffW> ? (I'll put myself in the ?-queue to keep things fair.)
.Dan Eicher> After looking at the spec's in BYTE on DEC's new Alpha
processor, one would think that all other processors just
become obsolete, have you reviewed the Alpha proc. and do
you have any thoughts on it and its impact?
.KarlG> I have only briefly looked at DEC's Alpha (ironically the code
name for the 99000 at TI was ALPHA). There are a number of
factors that determine the best processor, such as speed,
cost, software base, etc. The processor I'm currently working
would require a general purpose processor with over 2 Billion
operations per second to keep up with it AT IMAG PROCESSING.
ga
.Don O'Neil> ?
.JeffW> ga, Michael
.Michael.> What in the form of tech info and programming info, aside
from the databook, is out there on the 99105? and where
can I get it?
.KarlG> Unfortunately, the product is basically obsolete at TI, and
there is very little else other than the data manual that you
can get from TI. ga
SHH> ?
.Dan Eicher> ?
.JeffW> Some of our may be interested in why the 99/4A is as it is,
with the 8-bit bus and interpreter GPL. Could you tell that
story? Maybe even backup to the 9985?
.KarlG> Yes. A few questions back I mentioned the 9940. Well, even
though the 9940 was a disaster of a design, it was spun off
into a a chip called the 9985. The 9940/9985 was achitected
with an 8-bit data path -- essentially it was an 8-bit
processors emulating a 16-bit processor (the problem was it
took almost as much hardware as a full 16-bit design so it
cost too much and had lower performance).
.KarlG> Meanwhile, the home computer group was "ordered" (by the then
TI president J. Fred Bucy -- or at least that is how the
story goes inside TI) to use a TI processor. So the 9985 was
chosen (it may have been that the 9985 coversion was done
expressly for the home computer). But as I say, the 9940 and
thus 9985 didn't work! But .... the home computer group had
already designed the home computer around the 9985. So in
the hope that eventually the 9985 would be fixed, they took
the 9900 and ADDED hardware around it to make it look like an
8-bit bus 9985 -- thus they paid for a 16-bit processor and
hobbled it to an 8-bit bus. BTW, you may also note that the
9985 was to have 256 bytes of SRAM on-chip, and that is the
RAM that was put next to the 9900 (all other RAM on the 99/4
was off on the other side of the 9918 (and thus slow to
access). Regarding GPL, the story goes that what the home
computer designers really wanted was their own, more Z80 like
CPU that would run an opcode set called GPL (originally Games
Programming Language, later re-cchristened Graphics
programming language). Well ... in hope (boy they had a lot
of hope) that they would eventually get the CPU of there
dreams (ala one that executed GPL directly), they interpreted
EVERTHING into GPL. THUS when you ran BASIC, a BASIC
instruction is fetch from the 9918's memory over an 8-bit bus,
then converted into GPL, then converted into 9900 assembly
code. The other engineers at TI at the time used to say, "The
TI has the best bag of parts of any home computer, they just
put them together wrong"
.KarlG> Some of the above, may be apocryphal, but it is as I saw or
heard of it. ga
.JeffW> ga, Don
.Don O'Neil> Do any of the logic diagrams still exist for any of the
9900/99000 series chips? Is it possible to get any copies
of these?
.KarlG> I have the logic diagrams for the 9900, 9995 and 99000 in my
office, but we/TI do not generally give that kind of
information out... BTW, the 9995 was originally called the
9985A but it was several times faster than the 9985. I
designed it with a 16-bit bus to be much faster and smaller
than the 9985 and it eventually was called the 9995. ga
.JeffW> ga, SHH
SHH> could you briefly tell me what the limiting factors are for
realistic tv like movement and realism , and do you see it
available soon for pc's soon?
.KarlG> Over the next few years I think a destop system will be able
to provide some fairly realisted simulation graphics
(equivalent to a high end workstation today). I expect that
we will see about 500K fill trangles per second drawing speeds
(today's PCs are at perhaps 1 to 10K). We will have two way
video confererencing capability. The system will be able to
read a greater than a page a second any typical printed page.
The technology is already being developed, and the chips will
come out over the next couple of years. It may take about 5
years before it is commonplace. ga
.JeffW> ?
.JeffW> ga, Dan
.Dan Eicher> Karl, TI since the days of calculators with LED's had
been a leader of Advanced handheld calculator I think the
line was ti-85, cc-40, ti-74, procalc-95. Then something
happened at TI. No NEW calculators that push the leading
edge have been introduced. It seems TI has relinquished
the leadership position to HP, Sharp and Casio. TI seems
happy pumping out millions of low dollar/medium
functionality calculators. Do you think someday TI will
reenter that market as a performance leader?
.KarlG> TI has focus their efforts on more educational calculators,
such as one that handles fractions (ie it can add 1/3 plus
3/4ths). We still make scientific calculators as well. The
calculator market as a whole is not that big a business when
compared to the PC and rest of the electronics industry, which
may explain a lack of focus on high end calculators. Also for
the most part calculators are by today's standards low tech
products.
.Dan Eicher> ?
.JeffW> Could you tell us more about the revolutionary imaging
processor you are currently working on developing?
.KarlG> I can't say too much about it yet. There will be an overview
article about in in the November IEEE Computer Graphics and
Applications (I'm also guest editing the issue along with Jack
Bressingham, of line drawing fame, and Nick England, from
SUN). The chip is capable of over 2 billion RISK-Like
operations per second and is targeted at image processing (it
is meant to be like TI's DSPs/320 chips only targeted at
imaging). What is both interesting (and a little scary 1000
times faster than a 34020. ga
.JeffW> ga, Dan
.Dan Eicher> Karl, yesterday, while digging around at a ham fest, I
came upon a chip labled TMS9996, having memorized, alarge
portion of the 99xx databooks, I took this chip home to
look for it.... I can't find it! Have you heard of it?
.KarlG> I think that might be a 16-bit bus version of the 9995. The
9995 had a 16-bit internal bus (that when to 256-bytes of
internal 16-bit wide RAM) but brought out an 8-bit external
bus (ala the 8088 or 68008). It may also have been a special
chip for supporting our hardware emmulator (or both). ga
.JeffW> Since there is no one in the ?-queue, I'll ask another.
.Don O'Neil> ?
.JeffW> Did any other major computer manufacturer ever consider using
a 9900 family CPU in a personal computer?
.Dan Eicher> ?
.KarlG> I really don't know. The 9900 was used in a number of
products other than the home computer (note it was about 4
years old when the home computer came out), but I am not very
familiar with its other uses. ga
.JeffW> ga, Don
.Don O'Neil> What are the current pixel height/width/depth limitations
of the 340x0 series processors? What's in store for the
next 340x0 processor? Why the 010, 020, 030 suffixes? Is
there any link to the Motorola suffix(industry rumour).
.KarlG> There are essentially no resolution limits of the 340
architecture. It can handle 1,2,4,8, 16, or 32-bits per pixel.
Other than some speed enhancements of the 34020, there are no
plans for a 34030. The suffix is just TI's way of numbering
follow-on family members, it may have been patterned after
Motorola's, but I can't say -- Intel for example has kept with
their "lucky" 86 nomenclature and has used the middle digit to
signify generations. TI's next step in the graphics/imaging
market will be more revolutionary than evolutionary. ga
C_BOBBITT> ?
.JeffW> ga, Dan
.Dan Eicher> Do you know, what kind of deal was struck between Yamaha
and TI that allowed them to continue development of the
9918. Aka... The 9938, 9958 ect. If you where on Yamahas
9938 design team, would you have done anything different?
.KarlG> As far as I know, there was no deal. TI was working on a new
follow-on to the 9918 that we also called the 9938 (code name
was AVDP). But the video game market collapse didn't
materialize (the group doing videotext went on to found
Prodigy), and the AVDP design had a lot of problems. Faced
with a falling market and a design with problems TI stopped
the program. My understanding is the Nintendo is based on a
9938 clone and that the Yamaha 9938 was a 9918 clone (I
understand they emulate all the 9918 registers etc.) ga
.Dan Eicher> c: Nintendo has the same problem with sprites
"dissapearing" if more than 5 per line...
.JeffW> ga, Chris
C_BOBBITT> Lately there has been a lot of interest in a 99/4A..
emulator that runs on PC compatibles. I've heard that the
latest.. version runs about full speed on a 386-40Mhz. Is
there any processor on the... market today that would be
fairly amenable to running a 9900 emulation? ga
.KarlG> It is hard to emulate any processor exactly on any other
processors. A rule of thumb is that the same function will
run about 100 times slower. It can also be very difficult to
emulate in software. Thus, it is often cheapest and easiest
to just use the original processor. ga
C_BOBBITT> what if it isn't readily available, like the 99105?
<smile>
.JeffW> c: Newark Electronics will sell you a 99105, Chris.
.Dan Eicher> c: or a thousand!
.Dan Eicher> ?
.JeffW> ga, Dan
.KarlG> I think you can still get the 99105, but I'm not sure. You
would have to run one of today's fastest processors to get
close to emulating a 99105 in "real time". ga
.JeffW> ?
.Dan Eicher> The 9901 is a neat chip, but is getting hard to find. Do
you have any pointer on designing one with perhaps two
pals and the prologic compiler?
.KarlG> I don't know, The 9901 was a pretty simple chip. It was also
a poor combination in that it combined the interrupt
interface, of which you need 1 per system with I/O ports that
you may need many per system (ie people ended up with mulitple
interrupt controllers they did not need to get enough I/O
ports). ga
JERRYC> ?
.JeffW> I was looking around for a possible functional replacement for
the 9901. I noticed that Motorola makes an 8 interrupt, 8
parallel I/O, and 4 timer chip. Wonder if that -901 suffix is
coincidence? Er, chip number is 68901.
.KarlG> I think it is purely coincidental, The 9901 was based on the
9900's CRU interface which was specific to the 9900 family
.KarlG> --- How about a few last questions and rapping things up ---
.KarlG> ga
.JeffW> ga, Jerry
JERRYC> I heard that the 340x0 designs profited from coordinated ram
chip designs -- is TI pursuing this approach further?
.KarlG> Yes, I was one of the original definers of the Video RAM that
was used with the 340. I am now very involved with TI's new 4
Megabit Video RAM. It can be a big advantage when designing a
processor to know (and have input into what the RAMs will look
like when the processor comes out). ga
.Michael.> ?
.JeffW> ga, Michael
.Michael.> The CRU addressing scheme is an interesting design, I
wonder If you could reflect on it some more.
.KarlG> In hindsight, many at TI thought is was a mistake. The way
CRU worked, the whole CPU and it's address bus gets tide up
and it only sends 1 bit of data per I/O cycle. From a design
point of view, the CRU instructions were amoung the most
complicated instructions to microcode (the code inside the
chip that interprets the instructions). Most of us felt that
it would have been much better to have put a serial shift
register that the CPU could load in one cycle to shift off the
chip, and another serial register for shift on the chip
(essentially was most microcomputer chips have today). ga
.Michael.> ?
.JeffW> ga, Michael (this will be the last question taken tonight)
.Michael.> What drove TI to use the reverse convention on the
address and data bus, ie A0 is the MSB instead of the
LSB?
.KarlG> In the days of that decision, IBM used "Big Endian" (MSB=0),
and DEC used little endian (LSB=0). TI choose to follow IBM.
Intel followed DEC and Motorola's 68000 was all mixed up (they
numbered their bits LSB=0 but ordered it. the bytes so that
the most significant byte was 0). It turns out that both
floating point and graphics are more "naturally" big endian,
and integer operations are more naturally little endian.
Unforturnately it is a big mixed up world (PC's are little
endian, MAC are basically big endian). Today many processors,
including the MIPS, 34020, and some other new processors can
power up to be either big or little endian. By the way the
9918 was originally designed in "little endian" and at the
last miniute, to be 9900 compatible is was switched to big
endian (which is why the address bus may appear to be kind of
funny in how it connects up when using different size RAMs).
ga
.JeffW> Thanks for being our guest tonight, Karl. If you have any
questions about using Delphi, let us know.
.Michael.> Thanks Karl, Jeff. Great CO tonight.
.KarlG> I enjoyed it too
.JeffG> Karl, as System Manager of TI NET, I would like to thank you
for taking the time tonight and share with us your knowledge.
Forthe Staff of TI NET, we thank you and hope you'll be able
to join us again at a future time. We appreciate it. Jeff
Guide
JERRYC> Thank you Karl!
.KarlG> I appreciate that, now it time to get to bed!
.JeffW> Thanks again, Karl. And if you have any questions on what is
happening in the 9900-world today, Delphi is the place to find
out.
.KarlG> bye
.JeffW> Goodnight.
.KarlG> - signed off -
------------------------------------------------------------------------
This conference transcript was edited to make the reading of it
easier to follow. No significant material was removed or edited.
TI NET welcomes BBS's, Networks, and newsletters to use this
transcript and the following without permission, provided that BBS's
upload it in its entirety, without adding to or editing this transcript.
Newsletters may use excerpts. As a courtesy, all users should credit "TI
NET on Delphi."
TI NET FREE SIGN UP OFFER
Dial, via modem:
1-800-695-4002. Press Return once or twice. At the
Password prompt, type CUSTOM. When asked for the referral code, enter
TELEDATA
Pick the custom payment plan best suited to your budget and
requirements.
Here is signup information for Delphi, which has the TI International Net SIG
visited by many notables in the TI 99/4A community.
Cambridge MA, December 9, 1992 -- DELPHI, an international online
service, now provides full access to the Internet including real-time
electronic mail, file transfers with "FTP," and remote log-ins to other
Internet hosts using "Telnet." With this announcement, DELPHI becomes
the only leading consumer online service to offer such a wide
variety of Internet features.
Russell Williams, DELPHI's general manager, explains the significance of
this announcement. "Because DELPHI is a well known consumer online
service, the Internet will now be easily accessible to much broader base
of personal computer users. DELPHI has easy-to-use menus, low access
rates, and over 150 services in addition to the new Internet gateway.
DELPHI is also easy to reach; there are access numbers in over 600 cities
and towns throughout the US and in many other countries. The combined
resources of DELPHI and the Internet is a great package for anyone
interested in online services."
The Internet is considered the world's largest computer network. It is
comprised of thousands of companies, colleges, schools, government
agencies, and other organizations. There are currently an estimated 4
million users. "This incredible reach of the network will mean
better and more specialized services for all users" adds Mr. Williams.
"For example, users can take electronic courses conducted by leading
universities, access reports from government agencies, and
get product information and support directly from companies. There are
also mailing lists and discussion groups for almost every special
interest imaginable. Electronic mail can be used to send private
messages to anyone on the Internet and even many commercial networks like
Compuserve and MCI Mail."
DELPHI's connection to the Internet works both ways: In addition to
offering access out to other networks, DELPHI provides value-added
services to people already on the Internet. Any user of the Internet
can access DELPHI to use services such as Grolier's Academic American
Encyclopedia, the Dictionary of Cultural Literacy, Reuters and UPI
newswires, stock quotes, computer support, EAASY Sabre travel
reservations, special interest groups, real-time conferencing,
downloadable programs, and multi-player games. All these services can
be reached through the Internet simply by joining DELPHI and then
telnetting to the address "
delphi.com" via the commercial Internet.
In order to help new users with questions related to the Internet,
DELPHI has an area online to provide support. The Internet Special
Interest Group (SIG) includes an active message forum where members and
staff can exchange useful information. Comprehensive guide books,
downloadable software, and information files are also available.
DELPHI has two membership plans: the 10/4 Plan is $10 per month and
includes the first 4 hours of use; additional use is $4 per hour. The
20/20 Advantage Plan is $20 per month, includes 20 hours of use, and is
only $1.80 per hour for additional time. Rates apply for access speeds
up to 2400bps; 9600bps access is currently being tested in a few locations.
The Internet service option is an extra $3 per month and includes a
generous transfer allocation of 10 megabytes (the equivalent of about
3,000 type-written pages). Access during business hours via Sprintnet
or Tymnet carries a surcharge.
Through a special trial membership offer, anyone interested in learning
more about DELPHI and the Internet can receive 5 hours of access for free. To
join, dial by modem,
1-800-365-4636 (current Internet users should telnet
to "
delphi.com" instead). After connecting, press return once or twice. At
the Username prompt, enter JOINDELPHI and at the password prompt, type
INTERNETSIG. DELPHI Member Service Representatives can also be reached by
voice at
1-800-695-4005.
DELPHI is a service of General Videotex Corporation, a leading developer
of interactive and online services based in Cambridge, Massachusetts.
For further information, send email to Walt Howe, DELPHI Internet SIG
Manager (
walt...@delphi.com).