Why has T.I. consistently tended to produce faulty products throughout the company history?
This quote makes me think of many more:
"Texas Instruments didn't seem to put much effort into the layout, which Mazor calls 'pretty sloppy techniques' and 'throwing some blocks together'."
I recall the TI-99/4a that gave the TMS-9900 CPU a few hundred bytes of RAM, and prevented peek/poke or any other low-level manipulation of the machine (and trashed its educational value). I really wish TI had stayed out of this market.
They sold 386 Xenix machines for a while, and their motherboards came in three pieces connected by ribbon cables, which did not do much for reliability compared to Compaq. They should have used a one-piece motherboard design.
Later 68000 NuBUS TI-UNIX systems were developed when MIPS and ARM were both visible. What were they thinking?
The DSPs and other core competencies must be quite good to cover for these kinds of mistakes. Their SPARC chips never seemed to get any complaints, but I would be curious to hear Scott McNealy's take.
T.I.'s lobbying department is about ten times as skilled as their engineering department. Schools don't embrace the TI-83 calculator because it's the best, most economical tool for the job, they do it because they're legally obligated to. T.I. has built empires on peddling electronics with thousand-percent markups.
The calculator business is a tiny part of TI. Wikipedia says less than 3% of revenue; that seems high to me but I don't remember a specific number from my time there. I can assure you, though, that approximately all of TI's engineering and business efforts are directed towards semiconductors. You will note, for instance, that the words "calculator" and "education" do not even appear in their second quarter financial statement[1], while the words "analog" and "embedded processing" do.
MIPS showed up in '85 but by then there were dozens of 68k based unix boxes in the field (I worked for UniSoft, we did over 100 68k unix ports, including the lisa and the mac 2.
Back in the mid-80s CISC was still a real thing, intel hadn't really embraced 32-bits yet, if you wanted a unix microprocessor you went for a motorola 68k or nat semi 32k (there were a couple of others with a lot fewer design wins) - at the time every man and his dog seemed to be building a 68k based unix box in his or her garage - we had people come from Yugoslavia, one group Sweden bought a 1st class seat for their wire-wrapped prototype ....
> Later 68000 NuBUS TI-UNIX systems were developed when MIPS and ARM were both visible. What were they thinking?
Inexpensive unix workstations for running established and embedded applications. Also, they developed a lot of chips and designs for certain other large 68k+nubus players, so it certainly wasn't just throwing money into a one-and-done pit.
BTW, whatever they were thinking was correct, because afaik those TI1500 machines were profitable for them.
They were also producing SPARC chips for Sun by this point.
Why on earth choose 68k when the entire market saw SPARC wiping the floor with CISC?
EDIT: If Wikipedia is right, Fujitsu made the original SPARC v7 (MB86900) chips starting in 1986; T.I. didn't start SPARC manufacture until 1992 with a SPARC v8 (SPARCStation 10).
Well, it was a good choice, because it made them money for that product line. So they must have known something (or gotten really lucky)
Like I said, existing 68k applications and userbase were a huge draw. Using existing development work to establish a product line. Time tested embedded designs over the newer and less long-lived SPARC.
These were never supposed to be cutting edge - that's a feature, not a bug, sometimes.
Realistically, ARM-1 tapped out on April 26th, 1995, just as Olivetti was buying Acorn. This was a year before MIPS and SPARC if I am reading the dates right.
T.I. could have acquired the ARM processor. They likely would have killed it.
At 25k transistors, a UNIX machine based on it would have been far more profitable than 68k.
You don't have your dates right. You are a decade too late.
1985, not 1995.
Also, once again, it is TAPED out, not "tapped" out. I am guessing you are not a native English speaker; these two words have different meanings and pronunciations.
Taped: /teɪpt/ -- rhymes with "aped" or "shaped".
Tapped: /tæpt/ -- rhymes with "apt" or "clapped".
The way to remember this is that the double consonant (here, -pp-) "insulates" the vowel before it from being lengthened or changed by the "-e" after it.
Examples:
kit → kitten, bit → bitten, flat → flattened, net → netted. Same vowel sound in all these pairs.
Compare with:
kit/kite, bit/bite, bat/bate, sat/sate, not/note. All these pairs have different vowel sounds because the E after the consonant changes the vowel before the consonant.
It actually blew the VAX away, and a lot of other major systems.
"(1984) RISC II proved to be much more successful in silicon and in testing outperformed almost all minicomputers on almost all tasks. For instance, performance ranged from 85% of VAX speed to 256% on a variety of loads. RISC II was also benched against the famous Motorola 68000, then considered to be the best commercial chip implementation, and outperformed it by 140% to 420%."
No but it created a lot of sales for 68k customers who needed something faster than a 68030 sun3 and a lot of FUD in a Unix industry that didnt know whether MIPS, Itanium, PowerPC or half a dozen other architectures would be good enough to fend off a rapidly improving Windows onslaught.
In the end Intel played both sides of the aisle and won. Fun time to be in the industry though.
The lack of machine language programming out of the box was deliberate and similar to Apple's App Store business model. If you wanted to be a serious developer for the platform you had to buy an expensive 32K upgrade and floppy drive. Then you could use the assembler and other languages.
The RAM thing was an architectural choice - you weren't really meant to use the 256 bytes of SRAM for your programs, you were meant to share 16K of RAM with the video processor and make all read and write requests via the video chip. (You could access the 32K expansion directly)
I'd say the 99/4A was unique among the 8-bit computers. All the rest had basically the same architecture, but the TI had that strange system where most of the RAM had to be accessed through in's and out's.
Yes, the same architecture. You know what PaulHoule meant, and if you didn't you should have.
It's a fact that
* It is relatively easy to port between computers with the same 8-bit CPU
* It is harder, but still quite possible, to port between the Z80 and 6502
* Very, very little software was ported to the 99/4A, and until the end the lack of software was a major issue for a computer that sold almost 3 million units
The 99/4 architecture by and large baffled developers. Partly because of the unique CPU, but also because of fundamental design differences. I know of no Z80 or 6502-based computer which abstracts things the way 99/4 does. There is no TI equivalent (I'm talking about deviations from the prescribed norm, not the functionality per se) to the Atari 8-bit and TRS-80's multitude of DOSs, the C64's fast loaders, the Atari 8-bit's Happy Drive, Star Raiders and Eastern Front on Atari, Pinball Construction Set and Prince of Persia on Apple II, GEOS on C64. Since you're British, I'll add 3D Monster Maze on ZX81, Elite on BBC Micro, The Lords of Midnight and Knight Lore on Spectrum. Even the Color Computer, benighted by the 6809 CPU almost as unfamiliar to Z80 and 6502 developers as TMS9900, birthed Dungeons of Daggorath.
The App Store model only works if developers want to build apps for your system. In the microcomputer market, even IBM couldn't afford to make that assumption. Every successful vendor had to evangelize and grease the skids for developers as much as possible.
So the video chip functioned as a sort of memory controller? I've heard of people using modern GPUs to store arbitrary data. But that was just because it's there, it's fast and it is not in use most of the time. But this is different.
It probably isn't as crazy as it sounds. Graphics consumed a considerable amount of the memory on early personal computers. It looks like one of those video modes could consume about 75% of the 16 kB. Of course, that is cherry picking the most memory hungry mode. Other modes would use less memory, leaving unused memory for programs to use. If anything, it looks like the TI-99/4A is different since it appears to be designed for programs to use video memory because it's there and offers very little dedicated memory for the CPU.
Yes, that chip was the TMS9918A and it was also used in systems like the MSX computers and the Colecovision game console. The Colecovision was similar in that it offered 1KB of "scratchpad" RAM the CPU could directly address, and 16K of video RAM shared with the VDP.
The chip used auto-incrementing address registers that would allow a CPU to read from and write to VRAM.
“Texas Instruments, typical of their modus operandi in that era, had the most over-engineered solution. Instead of all-in-one chips, TI offered separate ICs for scoring, various types of graphics (walls, balls, stick figures, cars, rockets, etc.), game logic chips and position generators. You combined them together to make a complete game, which was fairly flexible for circuit designers but caused an expensive parts count for manufacturing, and the chip line ended up being a flop.”
That was supposed to compete with a single chip solution.
> The DSPs and other core competencies must be quite good to cover for these kinds of mistakes.
TI is primarily a semiconductor company. The consumer electronics business was sold off in the 90s. From what I've heard, I think laptops were the main product at that point.
I've used one of their ARM/DSP combos about 10 years ago. It was also just a bunch of random blocks thrown together (some video encode/decode, CPUs, DSPs). The interface between the ARM and the DSP was pretty slow, didn't seem well thought out. The SDK was written by some offshore team that didn't seem to really know the product that well and wasn't of the highest quality. Documentation was a bit iffy. In the end we managed to get our application to work. I thought that sort of quality/chaos issues were a more recent thing but I guess no?
The TI video chip on the other hand, the 9918, was foundational and was both copied, cloned and extended and powered millions of home computer, consoles and arcade machines. Even the NES was said to have been derived from the 9918 design.
Ken, could also throw into the list of "proto-micros" the General Instrument / Pico designs done in Scotland (1970) and mentioned in a few "real history of the microprocessor" articles over the years.
44 comments
[ 2.4 ms ] story [ 95.5 ms ] threadThis quote makes me think of many more:
"Texas Instruments didn't seem to put much effort into the layout, which Mazor calls 'pretty sloppy techniques' and 'throwing some blocks together'."
I recall the TI-99/4a that gave the TMS-9900 CPU a few hundred bytes of RAM, and prevented peek/poke or any other low-level manipulation of the machine (and trashed its educational value). I really wish TI had stayed out of this market.
They sold 386 Xenix machines for a while, and their motherboards came in three pieces connected by ribbon cables, which did not do much for reliability compared to Compaq. They should have used a one-piece motherboard design.
Later 68000 NuBUS TI-UNIX systems were developed when MIPS and ARM were both visible. What were they thinking?
The DSPs and other core competencies must be quite good to cover for these kinds of mistakes. Their SPARC chips never seemed to get any complaints, but I would be curious to hear Scott McNealy's take.
The calculator business is a tiny part of TI. Wikipedia says less than 3% of revenue; that seems high to me but I don't remember a specific number from my time there. I can assure you, though, that approximately all of TI's engineering and business efforts are directed towards semiconductors. You will note, for instance, that the words "calculator" and "education" do not even appear in their second quarter financial statement[1], while the words "analog" and "embedded processing" do.
[1] https://investor.ti.com/news-releases/news-release-details/t...
A lot of UNIX vendors had 68k machines. I was impressed in 2004 how smooth Domain OS/Aegis run on a 68020.
ARM-1 tapped out in 1985. No excuse really.
Back in the mid-80s CISC was still a real thing, intel hadn't really embraced 32-bits yet, if you wanted a unix microprocessor you went for a motorola 68k or nat semi 32k (there were a couple of others with a lot fewer design wins) - at the time every man and his dog seemed to be building a 68k based unix box in his or her garage - we had people come from Yugoslavia, one group Sweden bought a 1st class seat for their wire-wrapped prototype ....
These followed the TI 1300 386 Xenix machine.
Steve Jobs would have used them to buy ARM... or not, as NeXT was also 68k. I guess nobody saw what was going to happen.
https://www.home.unix-ag.org/engel/ti/s1500general.html
(Meaning the circuit design was stable.)
"Tapped out" means "surrendered, gave in, indicated that the fight was finished."
Inexpensive unix workstations for running established and embedded applications. Also, they developed a lot of chips and designs for certain other large 68k+nubus players, so it certainly wasn't just throwing money into a one-and-done pit.
BTW, whatever they were thinking was correct, because afaik those TI1500 machines were profitable for them.
Why on earth choose 68k when the entire market saw SPARC wiping the floor with CISC?
EDIT: If Wikipedia is right, Fujitsu made the original SPARC v7 (MB86900) chips starting in 1986; T.I. didn't start SPARC manufacture until 1992 with a SPARC v8 (SPARCStation 10).
Well, it was a good choice, because it made them money for that product line. So they must have known something (or gotten really lucky)
Like I said, existing 68k applications and userbase were a huge draw. Using existing development work to establish a product line. Time tested embedded designs over the newer and less long-lived SPARC.
These were never supposed to be cutting edge - that's a feature, not a bug, sometimes.
T.I. could have acquired the ARM processor. They likely would have killed it.
At 25k transistors, a UNIX machine based on it would have been far more profitable than 68k.
https://en.wikichip.org/wiki/acorn/microarchitectures/arm1
1985, not 1995.
Also, once again, it is TAPED out, not "tapped" out. I am guessing you are not a native English speaker; these two words have different meanings and pronunciations.
Taped: /teɪpt/ -- rhymes with "aped" or "shaped". Tapped: /tæpt/ -- rhymes with "apt" or "clapped".
The way to remember this is that the double consonant (here, -pp-) "insulates" the vowel before it from being lengthened or changed by the "-e" after it.
Examples:
kit → kitten, bit → bitten, flat → flattened, net → netted. Same vowel sound in all these pairs.
Compare with:
kit/kite, bit/bite, bat/bate, sat/sate, not/note. All these pairs have different vowel sounds because the E after the consonant changes the vowel before the consonant.
"(1984) RISC II proved to be much more successful in silicon and in testing outperformed almost all minicomputers on almost all tasks. For instance, performance ranged from 85% of VAX speed to 256% on a variety of loads. RISC II was also benched against the famous Motorola 68000, then considered to be the best commercial chip implementation, and outperformed it by 140% to 420%."
https://en.m.wikipedia.org/wiki/Berkeley_RISC
In the end Intel played both sides of the aisle and won. Fun time to be in the industry though.
TI-99/4a is what you get when you design in a complete vacuum, I mean ignoring the competition at the time.
http://shawweb.myzen.co.uk/stephen/tihistory.htm
GPL? GROMs? What on earth were they doing?
https://www.unige.ch/medecine/nouspikel/ti99/groms.htm
https://www.unige.ch/medecine/nouspikel/ti99/gpl.htm
Well Apple-II had its "Sweet16".
This one through me for a loop until I clicked through all your links!
The RAM thing was an architectural choice - you weren't really meant to use the 256 bytes of SRAM for your programs, you were meant to share 16K of RAM with the video processor and make all read and write requests via the video chip. (You could access the 32K expansion directly)
I did find the instruction set to be nice to code for once I'd upgraded enough to get access to assembly.
Z80 (Tandy, Sinclair, Amstrad, Elan, Memotech, etc.) 6502 (Acorn, Commodore, Atari, Oric, etc.) 6809 (Tandy, Dragon, etc.)
Some with a "GPU" that generated screen output, some without.
Some with sound generators, some without.
Some with disk interfaces as standard, some without.
Some with standard ports (Centronics parallel, RS-232 serial, IEEE, etc.) some without.
The same architecture?!
Yes, the same architecture. You know what PaulHoule meant, and if you didn't you should have.
It's a fact that
* It is relatively easy to port between computers with the same 8-bit CPU
* It is harder, but still quite possible, to port between the Z80 and 6502
* Very, very little software was ported to the 99/4A, and until the end the lack of software was a major issue for a computer that sold almost 3 million units
The 99/4 architecture by and large baffled developers. Partly because of the unique CPU, but also because of fundamental design differences. I know of no Z80 or 6502-based computer which abstracts things the way 99/4 does. There is no TI equivalent (I'm talking about deviations from the prescribed norm, not the functionality per se) to the Atari 8-bit and TRS-80's multitude of DOSs, the C64's fast loaders, the Atari 8-bit's Happy Drive, Star Raiders and Eastern Front on Atari, Pinball Construction Set and Prince of Persia on Apple II, GEOS on C64. Since you're British, I'll add 3D Monster Maze on ZX81, Elite on BBC Micro, The Lords of Midnight and Knight Lore on Spectrum. Even the Color Computer, benighted by the 6809 CPU almost as unfamiliar to Z80 and 6502 developers as TMS9900, birthed Dungeons of Daggorath.
OK. I guess.
This is not any standard definition of "architecture" that I recognise, but I think I see what you are getting at.
The chip used auto-incrementing address registers that would allow a CPU to read from and write to VRAM.
“Texas Instruments, typical of their modus operandi in that era, had the most over-engineered solution. Instead of all-in-one chips, TI offered separate ICs for scoring, various types of graphics (walls, balls, stick figures, cars, rockets, etc.), game logic chips and position generators. You combined them together to make a complete game, which was fairly flexible for circuit designers but caused an expensive parts count for manufacturing, and the chip line ended up being a flop.”
That was supposed to compete with a single chip solution.
TI is primarily a semiconductor company. The consumer electronics business was sold off in the 90s. From what I've heard, I think laptops were the main product at that point.
https://web.archive.org/web/20160323184954/http://www.ti.com...
The Texas Instruments TMX 1795: the first, forgotten microprocessor - https://news.ycombinator.com/item?id=9520210 - May 2015 (30 comments)
http://www.pat2pdf.org/patents/pat4001566.pdf