There are rare computers and there are very rare computers, and there are computers that are extremely hard to come by. Recently, almost by accident, I was able to acquire one of the latter, one that I always (since the time you could actually buy it as a brand new product) wanted to have. It looks gorgeous, even nowadays, because it is basically a steel case where the colour is not ageing (and because it comes from a collector that obviously took very good care for this machine). But enough rambling, let’s look at my precioussss…
The Atari Transputer Workstation (also known as ABAQ, ATW-800, or simply ATW) was a workstation class computer released by Atari in 1989, based on the INMOS Transputer.
As some of you might remember, Transputers were considered to be the Next Big Thing in the late 1980s. Transputers wanted to solve the problem of increasing the performance of a computer system without the need of having to develop faster CPUs (which was already then considered to be economically feasible only up to a certain limit. This limit was reached in a way in 2001). Instead, an arbitrary number of cheap but complete CPUs should collaborate to provide the needed performance. Sounds familiar? Yes, its basically the same concept as multi-core machines today with the difference that Transputers were separate chips that also did not share caches. As the collaboration of CPUs was very important for this approach fast (for the time) interconnections between the Transputers were built into each of them that could extend even outside a single computer system and therefore connect multiple Transputer computers to a combined system. Transputers contained a built-in RAM controller, so RAM could be added easily.
Transputers were the product of a single British company, Inmos that released the first Transputer in 1985. Transputer systems could not hold up to their more traditional competition, and in 1989 Inmos was sold to SGS Thomson. After that, Transputers were basically discontinued.
Inmos designed these Transputer CPUs models in it’s lifetime:
17.5, 20 MHz
17.5, 20 MHz
with on-board disk controller
15, 20 MHz
20, 25, 30 Mhz
20, 25 MHz
64 bit floating point support
20, 25 MHz
64 bit floating point support
20, 25, 30 Mhz
64 bit floating point support
The ATW and its operating system, HeliOS, was conceived by Perhelion, a company that was founded by former employees of MetaComCo. As MetaComCo had good connections to both Atari and Commodore, Perhelion tried to interest both companies in releasing a Transputer workstation running HeliOS. Commodore had expressed some interest in their new system, and showed demos of it on an add-on card running inside an Amiga 2000. It appears they later lost interest in it. It was at this point that Atari met with Perihelion and work started on what would eventually become the ATW.
The machine was first introduced at the November 1987 COMDEX under the name Abaq. Two versions were shown at the time; one was a card that connected to the Mega ST bus expansion slot, the second version was a stand-alone tower system containing a miniaturized Mega ST inside. The external card version was dropped at some point during development. It was later learned that the “Abaq” name was in use in Europe, so the product name was changed to ATW800.
The ATW system came in a large tower case. It consisted of three main parts:
the main motherboard containing a T800-20 Transputer and 4MB of RAM (expandable to 16MB)
a complete miniaturized Mega ST acting as an I/O processor with 512kB of RAM
the Blossom video system with 1MB of dual-ported RAM
All of these parts were connected using the Transputer’s 20 Mbit/s processor links. The motherboard also contained three slots for additional “farm cards” containing four Transputers each, meaning that a fully expanded ATW contained 13 Transputers. The bus was also available externally, allowing several ATWs to be connected into one large farm. The motherboard also included a separate slot for one of the INMOS crossbar switches to improve inter-chip networking performance.
HeliOS was Unix-like, but not Unix. Of particular note was the lack of memory protection, due largely to the lack of an MMU on the Transputer. This is not quite the issue it might seem, as the Transputer’s stack-based architecture makes an MMU less important. Meanwhile HeliOS was Unix-like enough that it ran standard Unix utilities, including the X Window System as the machine’s graphical user interface (GUI). In addition HeliOS ran on all of the Transputers in a farm at “the same time”, which allowed all computing tasks to be fully distributed. Turning off an ATW would not affect the overall farm, the tasks would simply move to other processors on other systems. Later HeliOS was ported to other processors including the ARM architecture.
The Blossom video system was developed specially for the ATW. It offered 4 different video modes up to 1280 by 960 pixels at 16 out of 4096 colours. The Blossom also included a number of high-speed effects (128 megapixels/s fill rates) and blitter functionality, including the ability to apply up to four masks on a bit-blit operation in a fashion similar to a modern graphics processing unit’s ability to apply several textures to a 3D object. The team in charge of the Blossom would later work on another Atari project, the Atari Jaguar video game console.
There is an ATW price list in Pound Sterling (GBP) stating the prices for the machine and various options excluding VAT:
5000 GBP in 1990 equals to about 13700 DEM or 8000$ at the time which corresponds about 9900 GBP or 14000$ today. Quite a price… On the other hand, an Atari TT was 3000$ in 1990.
It took quite long in a PC before a machine could handle more than 4 processors or cores.
I also looked into how the ATW compares to other product-level Atari computers in terms of speed. MIPS-wise, a corresponding list looks like this:
3.84 MIPS (Motorola DSP: 16 MIPS)
8 MIPS (I guess because is runs at 2*16MHz)
10 MIPS (per T800, i.e. 130 MIPS for 13 T800-20)
One can argue that the DSP inside the Falcon has a quite hefty 16 MIPS, and that a combined 20 MIPS for the Falcon (CPU + DSP) is more than the combined 11 MIPS of the ATW, but first, a DSP is not a general purpose processor, so this power is not available to every program. Second, you could add up to 12 T800-20 inside an ATW… So, although the ATW did not run TOS, and it therefore not the fastest ST that has been sold by Atari, it was the fastest computer by Atari. Of course, later projects (e.g Hades) would have been much faster.If we look at the cost per MIPS, we can state the following:
cost per MIPS
no farm card
1 farm card
3 farm cards
So, if you needed to have compute power, a loaded ATW was an economic option.It is said that only between 200 and 350 ATWs have been built, out of which 50 to 100 were prototypes that were released already in in May 1988. The production run has been released in May 1989. Another rumour is that 200 ATWs were sold to Kodak. The label on the back of an ATW say something like:
Serial Number: AB84A 90XXXX
The serial numbers that I know are:
It says also: Made In Germany. That sounds unusual. It probably means that the ATW was assembled by a 3rd party.If you have ever heard of Transputers outside this text, it was probably a long time ago. This effect typically indicates that a technology was not successful as it is also the case here. For the ATW 800 there are three groups of reasons for the failure of this machine:
this machine was ways too pricey for the mass market
Atari seem not to have invested much time and effort in supporting this model or to develop successors (I can also imagine they made a loss on every machine)
HeliOS was a too exotic environment
Perihelion remained the exclusive distributor in England (and it was always a small company)
Transputers as a technology failed because they had problems in terms of pricing, and later on performance compared to the (traditional) competition
Inmos as the sole manufacturer of these CPUs was a too small company
finally, Inmos folded basically in the same year as the ATW was published
Still, despite the failure of the machine for the masses (:-)), the ATW 800 was a good computer and had the potential to be used advantageously in some niches like scientific computing. A running ATW 800 is still the best opportunity to experiment with the Transputer technology. If you can get one, that is. It is rare to a ridicule degree.Technical Data
CPU: Inmos T800-20 @20 MHz (10 MIPS)
RAM: 4MB (expandable to 16MB)
Graphics: Blossom video system with 1MB of dual-ported RAM, supporting
mode 0: 1280 by 960 pixels, 16 colours out of a palette of 4096 (including 16 true greyscales, on a monochrome monitor)
mode 1: 1024 by 768 pixels, 256 colours out of a palette of 16.7 million
mode 2: 640 by 480 pixels (2 virtual screens), 256 colours out of a palette of 16.7 million
mode 3: 512 by 480 pixels, 16.7 million colours
Interfaces: RGB component display interface
Contains: a miniaturized Mega ST with 512kB RAM with all its interfaces
Released: May 1989
Number of produced machines: between 200 – 350 (of which 50 – 100 were prototypes)
In Germany, the Atari ST was very popular among users of “serious” applications such as text processing, CAD, and equipment controlling. This stemmed not least from the fact of the early availability of an affordable high framerate (70 Hz), high resolution (640×400) black-and-white monitor, the SM124 (and its successors).
Now, the Atari was conceived as a game machine (and Atari was initially really surprised by the demand of the SM124), later on morphed into the form factor of a business machine (Mega ST), but, of course, never had the form factor suited for industrial, 19″ rack-capable usage.
That’s where two German companies, IBP and Rhotron, saw a market. They converted Atari STs into modules that could be fitted into 19″ racks and added standard bus interfaces and measurement modules that could be used by the STs.
This is one of these models, the 190ST from IBP.
IBP presented the first version of this family in 1988. It was a licensed Mega ST design that has been re-designed to fit on three Eurocards. These were packed into a 19″ module with most interfaces, using industry-grade connectors at the front.
The 190ST was offered with one of three possible bus options:
Additionally (and in contrast to the original Mega ST), the 190ST also provided a socket for a 68881 mathematical co-processor. Other additional, built-in goodies included:
1 Watt audio amplifier
battery-buffered realtime clock
buffered DMA interface
Midi with up to 126kps (gilded 9-pin Sub-D connector)
keyboard via V24 interface (Sub-D)
Watchdog that can be software-controlled
application software sockets can use either ROMs or battery-buffered RAMs
Funnily, there was no mouse nor joystick interface on a 190ST. But you could add both using a special keyboard from IBP that was connected via V24…
Later on, the 190STV30 was added to the family. It featured an additional V30 CPU (8 MHz) in order to allow MSDOS compatibility.
Finally, the 190ST020 offered a Motorola 68020@16 MHz processor. It was introduced in 1991 and started from 5330 DM. As the 68020 was only used as a 68000 replacement, the bus width was unchanged (i.e. 16bit).
I always wanted such a machine, partly because it is a real Atari ST clone, not only a re-packaged one, partly because my first job as a student worker was to implement some software on such a machine. And, of course it is a rare computer made in Germany. Therefore, I was really surprised to find one in the US, from a commercial used factory equipment provider. It is the most basic model (68000, GemDOS, 512 KB RAM), but nevertheless :-)
CPU: Motorola 68000@8 MHz or 68020@16 MHz, V30@8MHz as an option
RAM: 0.5, 2, or 4 MB
HDD: internally none, but can be added externally (e.g. as another module)
OS: GEMDOS or RTOS
Graphics: Standard Atari ST graphics
Interfaces: Centronics, DMA, keyboard, Midi, RS232, video, floppy disk
The Buy It Now price is 1500$. The machine looks nice. The listing has a good compilation on Cat information. If you want to have a nice Forth computer and a non-standard design vintage computer, have a look!
This mobile computer from 1991 is the second pen-based device (after Grid’s Gridpad from 1989), the first one with a separate pen (although it beats the Gridpad 2050 or Gripad SL by only a few months if any) and the first one from a big manufacturer. It is also one of the most expensive mobile computers with an initial price of a hefty $4795 (about $8200 in 2014 numbers). Finally, it is one of the few mobile computers designed in Germany (by NCR in Augsburg).
Operating systems-wise the device is very flexible because it came out at the right time. As a PC it can run MSDOS plus NCR’s proprietary “PenOS”, a.k.a. the software that let you use the pen as an input device to enter text in a DOS environment. From 1992 onwards it could execute also (Microsoft) Pen Windows. Finally, even GO’s PenPoint was available for this computer. Some people even got (PC-)GEOS running on it.
My machine (as far as I can see) has only DOS + PenOS. If you happen to have PenPoint for it, I would be very happy…
The model number of this device is 3125. According to some old NCR information (see links below) this makes is a member of the NCR 3000 family that spans from this tablet and a notebook all the way up to a multi-processor, 100,000 MIPS big iron computer system under Unix. As all these systems do not share the same architecture or even the operating system, that’s quite a stretch…
The design of the device is very sleek; it looks very streamlined and timeless. The pen for example is neatly contained in a small hidden compartment at the front. It was rewarded a “iF product design award 1992 – Best Of Category”.
There seems to be a successor to this model with the Model 3130 NotePad in 1992. The 3130 had a backlit screen and a 40 or 60MB HDD and comes with Pen Windows. The weight increased by a pound. The price was about $4000.
I am very unsure on the fate of this machine. It was expensive, not often mentioned in the news, and so I assume it was not very successful. Maybe it was also subject of the turmoil following the takeover by AT&T in 1991/1992.
CPU: 80386SL @ 20 MHz (has about 15 MIPS)
Weight: 1500 grams
Display: LCD 640 x 480, 16 gray shades
OS: MSDOS plus PenOS or PenPoint or PenWindows
Interfaces: VGA, keyboard, RS232C, Centronics, all via a “I/O Connector Adapter”
In the insightful series of confronting today’s children with oldstyle technology, today: Kids react to Old Computers. Very funny and thought-provoking. Why did we accept that crap at that time? The answer is, of course, because nothing better existed at that time at that price point.
The WD-900 is the main board of a very rare computer whose CPU can execute P-Code (compiled from Pascal) directly. Before I come back to this computer please allow me a short detour :-) .
There are few computer architectures that aim at executing code that is closer to a certain programming language directly on the CPU. And none of them were successful in the sense that they sold to a larger extend because simply the technical progress on CPUs that do not have to obey such restrictions is faster than for these special CPUs. As a result, executing the programming language on general CPUs of the next generation is faster than doing it on the special CPU.
The list of programming languages for which such special CPUs exist(ed) is rather short:
Lisp (starting in 1975)
Forth (starting in the early 80s)
Prolog (starting in the 80s; research level only)
Java (starting in 1996)
Lisp machines did have some commercial success, but vanished in the early 90s.
Prolog machines never came out commercially although their development was one of the promises of the Fifth Generation projects.
Forth is considered by some not so much a high-level programming language, but something very close to computer hardware. There are still some interesting Fort CPUs products, so it’s probably more the low interest in Forth that leads to a low interest in Forth hardware (don’t get me wrong – I love Forth).
In contrast to that, Java is a high-interest programming language. Now, Java does not need to be executed directly on a CPU as it is often compiled into “Bytecode” anyway. Bytecode is a stack-oriented language like Forth. Of course, in contrast to Forth Java supports e.g. objects, but the principle is the same. Bytecode is a much simpler language than Java and better suited to be executed in hardware.
Now, the concept of Bytecode was not invented by Java. It existed long before Java, notably as the runtime system of UCSD Pascal, and now we are back at the WD-900.
In 1979, Western Digital, then a manufacturer of CPU and controller chips, looked for another use case of their MCP-1600 micro-coded, multi-chip microprocessor consisting of 3 types of chips:
CP1611 RALU – Register ALU chip
CP1621 CON – Control chip
CP1631 MICROM – Mask-programmed microcode ROM chip (512 – 22 bit words)
The main use of this CPU was as the processor in DECs LSI-11 computer, a compact, integrated version of the PDP-11 minicomputer. As the CPU was micro-coded and as the microcode was stored in one or more separate chips, it was easy to let the CPU execute a different command set by switching the microcode storage chips.
So what they did was to change the microcode to directly execute “p-code”, the bytecode of UCSD Pascal (of course, also p-code is a stack-oriented language). To that end they developed the WD-9000 chip set consisting of
CP2151 Data chip (was no different from the CP1611 of the MCP-1600 chipset and could be interchanged)
CP2161 Control chip
3 CP1631 MICROM chips
The difference was in the CP2161 control chip (and of course the MICROMs). Though the CP2151 contained multiple registers, but as the the p-code implementation was a pure stack machine, it did not use the registers.
In 1979, the competition were mainly 8bit machines. As a result, the MicroEngine outperformed e.g. a Z80-based machine at the same clock speed by almost a factor of 10. Of course, later 16bit machines like the 68000-based HP9836 (at 8 MHz, sold from 1981 for $11950) were faster by a factor of 3. Also, the performance advantage was eroded by the later availability of p-code to native machine code compilers.
The WD-900 board that I own is reportedly a New Old Stock board bought as a spare for a WD-90 computer that never has been used. It is boxed. The WD-900 board contains all the electronics: the CPU, RAM, serial interfaces and a floppy disk controller (WD1791/2) for two disks.
The WD-90 system contained a WD-900 board and a power supply. The (up to two) floppy disk drives needed to be attached externally.
The first boards shipped were poorly designed (power and ground traces the same size as signal traces, very few capacitors), required a large number of modifications, and even then did not work reliably. A couple of years would pass after introduction before a well-engineered MicroEngine was available. Between a damaged reputation and the introduction of the IBM PC, in the end the MicroEngine was not successful. You can see the lack of craftsmanship in the board design very clearly if you have a closer look on the photo of my board. Many patch wires, additional components and hand-soldering on a New Old Stock board…
The MicroEngine series of products was offered at various levels of integration:
WD-9000: five chip microprocessor chip set
WD-900: single board computer ($2995)
WD-90: packaged system ($5000)
SB-1600: MicroEngine single board computer
ME-1600: Modular MicroEngine packaged system
CPU: WD-9000@ 3.0 MHz
RAM: 64 kB (32k 16bit words)
Interfaces: 2 x RS232, 2 x parallel (i.e. floppy disk)
Released in 1994, the optimistically named “DeskTop Replacement 1” is an early pen-based, mobile computer. Like the NCR 3125 3 years earlier it’s a PC that you can carry in your hand and that you can operate using a pen as a mouse. Of course, the DTR-1 used updated hard- and software, but the idea is the same. Therefore, the architecture of these devices did not allow much freedom and required a desktop-class performance CPU. As a result, all these devices are the most heavy mobile pen computers with a weight almost twice as much as the one of an Apple Newton or a Magic Cap-based PDA. Not only were they heavy, the PC architecture also meant that the price was double or triple the price of a Newton or a Magic Cap device (a similar problem exist nowadays
in a lesser form for Windows-based tablets as opposed to Android-based tablets). The upside of the used PC architecture was that it sported all the standard interfaces also found on desktop PCs.
The machine ran on NiMH batteries. They were advertised to last for 3.5 hours. The pen ran on SR48 batteries and lasted for 350 hours.
The Dauphin DTR-1 could recognize handwriting and convert it to text on the fly.
The DTR-1 was manufactured by IBM.
The operating system was “Windows 3.1 for Pen Computing”.
Another very interesting feature about this computer is that it uses a tiny HP Kittyhawk 1.3″ harddisk. It seems to be the only computer where this drive came as a standard (it was an option in AT&T’s EO 440 Personal Communicator).
Of course, the DTR-1 was not a success (else this blog would not write about it :-) ). A quite steep price tag of over $2500 dollars where the initial Apple Newton costed only $700 a year earlier, a high weight, and an OS that was very exotic in the mobile market made the company starting to collapse in 1995. From the reported assets and debts, divided by the price for a DTR-1 I assume that Dauphin made at least 18000 units. Dauphin, however seemed to survived somehow at least until the year 2000.
The power supply of the DTR-1 is notoriously bad. People who own DTR-1’s recommend to use modern 12V DC power supplies instead of the original one. The original one is specified at 2.1A. The plug is center-positive. As a pen replacement old Fujitsu pens can be used.
In 1996 Dauphin also released a second model (called DTR-2) which was selling for $4445, but very few of them (in the few hundreds) seem to exist. The DTR-2 had a 486SLC2@50 MHz CPU, a 120 MB HDD, and 2 PCMCIA2 slots.
There are articles from 1999 about a “Dauphin Orasis” computer based on a Pentium@266 MHz, and there are people that report that they once had such a device, but these machines seem to be even more rare.
CPU: Cyrix 486SLC @ 25 MHz (has about 35 MIPS)
RAM: 4MB(expandable to 6MB)
HDD: HP 1.3″ Kittyhawk microdrive 40MB
Size: 5 x 9″
Weight: 1100 grams
Pen: active, requires batteries
Display: pen-sensitive, backlit, passive-matrix, monochrome VGA (640 x 480)
Interfaces: VGA (800 x 600, 256 colors), parallel and serial ports, Ethernet, Modem
Modem: Hayes-compatible (the modem and serial port are set to the same interrupt, so they can’t be used simultaneously)
Ethernet: the Ethernet module (apart from the connector) is an option