Wine

To get the obvious out of the way right at the beginning, the majority of the tools you will need to develop for the Jaguar are primarily maintained for Windows. I’ve found that they all run well under Wine. I installed it via brew.

The WINEDEBUG=-all environment switch is your friend.

Virtual Jaguar

The oldest Jaguar emulator still in semi-active development. Virtual Jaguar is developed and tested by many of the same people who maintain other key parts of the toolchain. Unfortunately compatibility could be better, with some titles and example code such as Atari’s 3D Demo failing to run at all. There is a macOS build of the last release (available here), but it does not display correctly on Sonoma.

BigPEmu

BigPEmu was originally developed to run the Jaguar games included in the Atari 50 compilation. Author Rich Whitehouse has since made the Windows version available for download, as well as recently porting it to iOS. Compatibility is excellent. The level of accuracy also seems equally high. The few exceptions I’ve found so far err on the side of improving the emulated experience over reproducing every annoying quirk of the hardware.

Noesis

Also written by Rich Whitehouse, Noesis is – among a great many other clever things which I’ve yet to investigate – a universal debugger. When used with the developer version of BigPEmu it allows full debugging of all three processors, with breakpoints, register views and memory inspection.

JagStudio

Built around ReBoot’s RAPTOR API, JagStudio allows 2D development in a dialect of BASIC, as well as C and assembly. Setting up a new project is still pretty low-level, requiring some knowledge of the format the Jaguar’s Object Processor is expecting data in, but overall if you just want to get started writing games this is the way to go. There are an enormous number of example projects included to learn from. Runs without issue on macOS using Wine.

jaguar-sdk

jaguar-sdk gathers together the modernised components of the Jaguar toolchain with documentation and example code in one repo. This is also made available as a Docker container, which works without issue on macOS. This is the method I ended up using. The example code included a 3D demo from Atari which formed the base of my game. The modified code is available in this repo.

U-235 Sound Engine

The U0235 sound engine provides sample and .mod file playback, as well as joypad reading and a random number generator. It is one of two sound options provided by JagStudio, and is simple to setup and use in your own project. Until an update is released, I suggest using the version from this comment, since it appears that the latest version available through the official web site isn’t the same bug-fixed version.

Additional Resources

• Updated Jaguar Software Reference The original Atari documentation, updated with the contents of both official tech notes and experience earned by the community over the years.

• Jaguar Programming Forum at AtariAge Decades of excellent information here. Expect to get sidetracked whatever specific topic you go looking for. Support for new questions about current libraries can be patchy.

• The Jaguar Underground Documentation Covers much of the same ground as the official documentation, with additional commentary and examples. Useful for a second opinion if something is the Atari documentation is unclear.

My project to build a Mistrum Spectrum clone is moving slowly, despite choosing to take what I’m sure will be an easier root. Lazarus Long has already done the hard work of translating the original schematics into a PCB design¹. I have had these manufactured and am slowly populating them. The issue of replacing the PROMs with something more modern – and more readily available – remains, but I think I’ve made some progress there.

Meanwhile much of my spare time has been spent absorbed in the Zen of assembly language programming, working on some retro game projects.

Tomb Robber is a Teletext mode Roguelike I started for the BBC Micro years ago. I haven’t worked on it recently. I mention it here so that maybe someday I’ll go back and finish it. I recall I was in the middle of replacing the combat mechanics when I last walked away from it.

Station Twenty One (provisional title) is a platform shooter for the Spectrum 128K. I chose to target the 128K partly to fit with my general wish to revisit all the old systems I once owned and write something for them, but mostly because the memory paging and multiple screen support allows me to easily do flicker-free graphics and thus appear competent. Having a proper sound chip is also nice. It’s just a shame that means it won’t run on the Mistrum.

Development is still in the early stages, but the main game mechanic is in place. All that’s left is to add some enemies and build out all the maps.

And because I can never have too many things on the go at once, I’ve also entered JagJam 2024². I never owned a Jaguar at the time, despite still being a full-on Atari fanboi when it was released. Looking at the prices on eBay it’s unlikely I’ll own one nowadays either. But the challenge of its unorthodox architecture³ was just too much to resist.

As with so many retro platforms, it’s amazing to see the effort put into maintaining usable dev toolchains. I’ve been playing around with Atari’s demo of the Jaguar’s 3D capabilities using the jaguar-sdk⁴. A tidied-up and rearranged version of this code is available on GitHub⁵.

Now to make a game from it.

Notes

Building modern reproductions of the ZX Spectrum is a staple of middle-aged retro-nostalgia YouTube. I’m a big fan. You can build according to the original design, but with (mostly) all new components:

Or you can build using fancy CPLDs as ULA replacements:

You can even build Soviet-style (discrete logic) clones:

Building a Spectrum clone looks like a piece of cake. So let’s have a go!

The Mistrum

At the beginning of 1989, the Czechoslovak Amatérské Radio magazine (issue 1/89¹) featured instructions and schematics for building the Mistrum, a ZX Spectrum compatible micro designed by Milan Prazan and Jaromir Mynarik. An English translation was later made available by Marek Bartoň². You can find a tidied-up version, with the diagrams restored and the errata from the next issue added, in the Archive section here. It’s a work in progress, which I’ll be adding to as the build progresses.

The introductory section of the Amatérské Radio article gives us a fascinating snapshot of our hobby as it was practised in Czechoslovakia at that time. The Iron Curtain still divided Europe. The Velvet Revolution would not begin in earnest until the end of that year. A cautious form of Perestroika is in place, but political and economic freedom is still absent. And yet here we are discussing how one would go about constructing a personal microcomputer – one styled after a Western design and most likely destined to be used to play games. The authors discuss how the Spectrum has become the most popular micro in Czechoslovakia – it’s cheap, simple, and there’s lots of software available for it – and the difficulties the builder will face in sourcing components, whether they look to the Communist Bloc or the Capitalist countries for parts.

The article continues with a section which presents an in-depth explanation of how the Sinclair Spectrum works, down to the timing level. The level of detail and the clarity of the discussion is excellent. It appears to draw from original research by contributors to the magazine over the previous years.

Next we come to the bulk of the article, the build instructions for the Mistrum.

The Mistrum is a clone of the Spectrum 48K with a few improvements. There is 64K of RAM, with the option to move the 16K ROM out of the address space so that it is all available, making running CP/M on the machine a possibility. There is support for keyboard-mapped joysticks. There is no extended memory banking, sound chip or serial ports, but as the authors describe the machine, it is designed as a base upon which each constructor can build.

The most interesting part of the design is the discrete logic replacement for the ULA. The authors name their replacement the ‘ULAM’. Faced with the problem of sourcing sufficiently-fast RAM for the lower 16K – Soviet memory was apparently not graded by speed – to allow for shared access by both the CPU and video circuitry, they took a different approach. An extra 8K SRAM is used to store the display data, to which the video circuitry has almost-exclusive access. Any writes to the 8K area of main RAM containing the display file are written through to this video RAM. One side effect is that all main memory can run without contention, which the authors claim yields an additional 3% performance over the Spectrum 48K. It will be interesting to see if this is the case.

What the Hell Am I Doing?

I have no idea what I’m doing. I’m a software engineer by trade. I know nothing beyond the very basics of electronics. I think I understand what’s going on with the logic gates, but there’s only so far that will get me. This will be a learning exercise. Why are all those resistors? Why?

There’s a temptation to assume that, because the Mistrum was designed to be built by Czechoslovakian comrades on their kitchen table with a box of scraps, the process will be easy. I’m going to try my best to avoid falling into that trap. I know it will not be simple, and I’m sure that the already high respect I have for those who pioneered microcomputing under such difficult circumstances will only rise higher. I will try to stick to the original designs as closely as I can – if for no other reason than I don’t know what I’m doing – but in some places it appears that will not be possible. Many components which were hard to source back then are even harder to find now. And I’m not going to even think about building my own power supply.

I am waiting for more components to arrive before I get started. I expect this is a method of procrastination I will employ a lot during this project. In the mean time there are already a couple of issues I will need to resolve:

PROMs: The original design uses PROMs to hold look-up tables for the video timings and address decoder. While some PROMs appear to be available from eg. AliExpress, there’s something about second-hand one-time-programmable components which I implicitly do not trust. I heard they’re also a pain to program. Reading around, it appears the best replacements would be either PALs/GALs or fast EPROMs. For now I will start with EEPROMs while researching SPLDs with a view to switching to those later if they seem a better solution.

Transistors: The bill of materials in the article lists most components by both their Western and local – generally Tesla – part numbers. The exception seems to be the transistors. Matching these to modern components appears to be a dark art. Luckily here I can lean on the work of z00m³, who has produced an interpretation of the Mistrum schematics which include the contemporary part numbers.

Layout: I’m going to start by building on breadboard. I expect I’ll have to change things around a lot as I learn, so it seems the easiest way to go. The article does include a suggested layout for the finished board, but half of it – page 33 – is missing from the linked Internet Archive copy. I’ll file this one under ‘it will make it more interesting’. While sets of full schematics are available online, I’m going to try to figure it out on my own, at least to begin with.

Resources

• Lazarus Long, who successfully built a Mistrum clone and provides a number of resources.

Notes

On April 20th 1982, Clive Sinclair announced the ZX Spectrum to the world. History doesn’t record the date on which he finally grew tired of the new machine, but it was certainly within that first year.

The actions of Sinclair the company follow closely the character of Sinclair the man, specifically through two distinct traits. Firstly we have Sinclair the inventor, naturally draw to the novel and innovative, but always with the eye of a marketer: the first, the cheapest. Even as it was being created, the Spectrum was far from Sinclair’s main focus. Jim Westwood, the company’s chief engineer and designer of the ZX81, was moved to focus on the flat screen TV project, leaving the relatively junior Richard Altwasser to work on the new machine. Lucky break that this was, it shows where the company’s priorities lay.

Secondly we have Sinclair the gambler¹. His game of choice was poker, a game of bluffing. It is clear from the chaos in the months following the launch of the Spectrum that Sinclair Research was not fully prepared. Production was not sufficient to meet demand. And yet the cheques kept rolling in and the company kept cashing them. Delivery within 28 days slipped back to first two months and then three. I can imagine that day-to-day life inside Sinclair during the remainder of 1982 was the kind of seat-of-your-pants environment which would appeal to the adrenalin junky gambler, with Clive called upon to regularly speak to the press and ensure potential buyers that everything was fine. And eventually it was. Everyone who ordered a Spectrum got one. Everyone whose Spectrum arrived broken got a replacement. The business of selling computers fell into a regular pattern. It got boring.

Sinclair’s relationship to the computers he’s best known for is at best ambivalent, even as they brought in the income to allow him to focus on other projects which he found more interesting.

I make computers because they are a good market, and they are interesting to design. I don’t feel bad about making them, or selling them for money or anything, there is a demand for them and they do no harm; but I don’t think they are going to save the world.²

“There weren’t projects within Sinclair to develop the Spectrum further,” Nigel Searle, Sinclair’s then Managing Director, would later admit³. The company would focus its efforts on the flat screen TV and the C5, and on the QL and other computing projects which weren’t the Spectrum. Outside Sinclair, however, things were different. An entire industry had been created around Sinclair’s computer. Software was written, hardware was built, and the machines continued to sell, both in the UK and in other countries.

Investrónica

César Rodríguez González⁴ was a Spanish emigre to Havana who made his fortune in the textile industry. He returned to his native land ahead of the Cuban revolution and settled in Madrid, where he already had business associates and investments. Among these was El Corte Inglés⁵, a tailors’ shop which would grow to become a chain of department stores across Spain and Portugal. With this growth came diversification. In 1980, Investrónica⁶ was formed as a wholesaler of technological products, focusing on textile manufacturing and consumer electronics. Among the products they brought to the Spanish market were the computers of Sinclair Research.

The ZX81 was a hit in Spain as it had been elsewhere. When the Spectrum was announced the interest from the public was there, but Spain would have to wait. There would be no Spectrums the Christmas of 1982. The launch would be delayed due to “technical issues”, Charles Cotton, Sinclair’s commercial director, told Investrónica’s director Ricardo García Gete⁷. As a last minute stop-gap, Investrónica struck a deal to import Dragon computers, but Dragon Data could not supply as many as the wholesaler would have liked.

The Spectrum would not come to Spain until April 1983, a year after its launch in the UK. When it did, it landed with a splash, announced through adverts aired during the wildly popular Un, Dos, Tres television show⁸. There were two spots, one focusing on the educational use of the Spectrum, the second on its use as a tool for balancing the domestic budget. Deals with banks across Spain reinforced this serious use and alone lead to 16,000 sales⁹. The Spectrum sold in great numbers, leading Sinclair to recognised Spain as its number one market outside of the UK.

And then… nothing. November 1984 saw the release of the Spectrum+¹⁰, the same 48K machine as was already available, but in a new case and with a new keyboard, to address one of the original’s main perceived weaknesses. Critics were not impressed. “At first glance it may look like a conventional professional keyboard, but a few seconds of use are enough to convince us otherwise.”¹¹ Investrónica would eventually offer the Spectrum+ with the ROM and keyboard localised to Spanish, going directly to Sinclair’s keyboard manufacture, Celluloid AB in Sweden, for the components¹². In part this was a move to stem the flow of grey imports from the UK¹³ which were undoubtedly cutting into their profits, as impatient users stopped waiting for the domestic market to catch up.

The launch of the Spectrum+ was accompanied by adverts touting the machine’s ‘64K’ of memory¹⁴, showing that maybe Sinclair had learnt something from their partnership with Timex – the same mathematical gymnastics as gave us the ‘72K’ TS2068. The market was evolving at breakneck speed, with newer, more advanced machines such as Amstrad’s CPC appearing. The TS2068 had disappeared along with Timex Sinclair, seeming to take with it the chance of bring its hardware evolutions to the main Sinclair line. The QL had been met with little interest by the Investrónica representatives who were given an early view of it¹⁵, and was showing typical Sinclair tardiness in terms of delivery. There were still a great many Spectrum fans in Spain, but Investrónica had nothing new to offer them.

128K

There does not seem to be a canonical timeline of the Spectrum 128K’s development available, despite numerous interviews with those responsible for its creation, both on the Sinclair Research and the Investrónica sides. The exact time, place, and reasons that development began are unclear.

The most widely cited inciting incident were two Royal Decrees issued during the summer of 1985. The first – no. 1250¹⁶ – announced on June 19th, covered “peripherals for input and representation of information in data processing equipment” and mandated, along with many other common sense requirements, that keyboards sold in Spain should include those symbols necessary to input the Spanish language. The second decree – no. 1215¹⁷ – announced on July 17th, was far more controversial. It introduced a tariff of 15,000 pesetas¹⁸ on all microcomputers imported into Spain. The Spectrum+, which was manufactured by Timex in Portugal and thus liable to the import tax, was selling for around 30,000 pesetas at the time¹⁹.

It is now commonly accepted that the second decree was the result of machinations by Eduardo Merigó²⁰, a former government minister and business man. His company Eurohard had purchased the assets of the now-failed Dragon Data²¹ and was in the process of moving its factory to Casar de Cáceres, Extremadura²². As Spain’s only domestic computer manufacturer, the resurrected Dragon would be free to take advantage of the home market without paying the tax. The decree did not go unchallenged by the existing import computer industry, and on August 28th, King Juan Carlos – on holiday in Mallorca at the time – signed decree no. 1558²³, which modified the previous decree to apply only to microcomputers with 64K of RAM or less. Famously this would lead to the Amstrad CPC472, a 464 with an extra 8K RAM chip rattling around inside, part of the machine but not available to the user²⁴. It is unclear who was behind this counter-move – possibly representatives of the newly-burgeoning IBM PC compatible market – but the change did not help Investrónica’s Sinclair business.

The Spectrum – still stuck at 48K – remained liable for the import tax and at a major disadvantage in the market. The question is whether this alone was enough to spur the creation of the 128K, or whether it came about as a more general response to Sinclair’s neglect of the platform. There are a few markers which may allow us to build a timeline. An early version of the Spanish version of the 128K’s ROM²⁵ – both the 48K BASIC and the 128K-specific parts – has been found and dated to April 1985. This is well before the announcement of either the Spanish language support requirement or import duty decrees. Even if we assume that an important industry player like Investrónica would have been forewarned of the upcoming changes, it would be surprising if they had enough notice to be this far along in development.

Rumours of the new 128K Spectrum²⁶ would reach the news stands just a couple of weeks before it was revealed at the SONIMAG trade show in late September²⁷. How close the machine shown there was to the one which would be released to the public that Christmas, or whether it was a fully-working pre-production prototype, is not clear. It wouldn’t be a Sinclair without some smoke and a mirror or two. But the timeline in comparison to the Royal Decrees seem very compressed: just one month after the ‘more than 64K’ reprieve, two since the original decree was announced. I think it is probably fair to say that the 128K was in development before the decrees, not in response to them. On the other hand, the pell-mell rush to bring the machine to market against these tight deadlines might suggests why the project was given the codename ‘Derby’.

Over the subsequent months more details trickled out. The machine would be manufactured in Spain, by Induyco – a fellow subsidy with Investrónica of El Corte Inglés – in Madrid, making it the second Spanish-built micro. Release in the UK would be delayed to the following year. No one was in any doubt as to the reason for this. Contemporary accounts made free mention of the glut of 48K Spectrums still unsold from the previous Christmas. Details of this over-production had been brought to light during that year’s chaotic search for funding to secure the future of Sinclair Research, with multiple potential saviours paraded in the press before bowing out.

The Spectrum 128K

What can we infer about the development of the 128K by looking at the final product?

Externally, the Spectrum 128K largely follows the design of the Spectrum+, using the same enclosure and keyboard as designed by Rick Dickinson, originally for the QL. Investrónica’s in-house industrial designer Guillermo Capdevilla is credited with adapting the case for the 128K²⁸. The most obvious feature of this tweaked design is the ‘toastrack’ heat sink, which internally is bolted to the voltage regualtor.

“The design of the [Spectrum 128K’s] ULA, designated 7K010E-5, is almost identical to the 5C and 6C ULAs used in the ZX Spectrum 16/48K,” Chris Smith tells us²⁹. The changes present are firmly geared toward increasing the quality of the video signal. The ULA now produces digital RGB, which feeds both a RGB monitor port as well as standard RF and composite, via a TEA 2000³⁰ video encoder. The system is driven by a 17.7345MHz clock to meet the demands of this encoder, and as a result the Z80 runs at a slightly faster 3.5469MHz. These changes seem like perfectly reasonable updates to the Spectrum’s video circuitry, evolutionary upgrades – like that of the keyboard – which could be seen as addressing the original model’s shortcomings, and not specifically geared towards the 128K. The only affordance for the new hardware appears to be the production of a clock signal from the ULA for the AY sound chip.

The major innovations of the 128K – the expanded memory, the sound chip, and the serial/MIDI port – are achieved via an innocuous little chip, the 10H8 PAL, an off-the-shelf component. It works with the ZX8401 and other discrete logic to implement the memory banking and IO port access to the AY-3-8912, through which the serial ports are driven. The ZX8401 is a custom chip made by Texas Instruments, which aggregates the function of a number of discrete logic chips in typical Sinclair cost-saving fashion, and appeared first on Spectrum 48K revision 5 boards.

There were rumours of Timex’s involvement in the design of the 128K. Certainly they had a relationship with Investrónica, who had been selling a version of their FDD 300 floppy drive system in Spain since late 1984³¹ and whose representatives had visited Timex’s Spectrum manufacturing facility in Portugal³². But there seems no evidence of their hand in the final machine. The improved ULA which they inherited from Timex Sinclair with its enhanced graphics modes – which was available in a PAL versions – is sadly absent. The memory banking scheme which the 128K uses is completely different from that employed in the TS2068. And while both systems use the same sound chip – which was very popular at the time and used by many other machines – they do via different IO port addresses³³. It’s a small detail, but it would have been a small change which would have promoted some degree of compatibility between Timex’s computers and the official Sinclair line.

The 128K RAM is divided into 2 physical 64K blocks and 8 logical 16K banks. Any of the logical banks can be paged into the highest 16K of the Z80’s address space (at $C000)³⁴. Bank 2 is always mapped at $8000, and bank 5 into $4000. The ULA can draw the screen from either bank 5 or bank 7, allowing for double-buffering. This means, however, that all banks which share that 64K block are contended, sharing access with the ULA³⁵.

The ROM is 32K, split into two distinct 16K parts – the 128K Editor and the 48K BASIC – either of which can be paged into the lowest 16K. The computer boots into the 128K Editor, which in the original Spanish version is the familiar BASIC prompt, but with a BASIC which has been re-implemented to require each keyword to be entered letter-by-letter, discarding the Spectrum’s unique keyword approach. It includes a rudimentary text editor which uses BASIC strings as storage³⁶. New keywords have been added to access the unique features of the machine, although unfortunately the extra 80K of memory is only available to BASIC in the form of a RAM disk. The PLAY command works with either the AY sound chip or a MIDI adapter plugged into the serial port³⁷. The SPECTRUM keyword takes the user back to the 48K mode for maximum compatibility with existing software, disabling the extended memory and paging-in the 48K BASIC ROM.

Sinclair would update the 128K’s ROMs in preparation for its UK launch. The text editor would be ditched in favour of a menu system, programmed by Rupert Goodwins³⁸, and a number of other bugs would be fixed³⁹. It would be released in January 1986. The first batch of machines were reportedly manufactured in Korea by Samsung⁴⁰, Sinclair presumably having fallen far out of favour with their UK manufacturing partners Timex, Thorn⁴¹ and A.B. Electronics⁴². Software houses were seeded with the machines pre-launch to ensure that a supply of software which made use of the sound chip and extra memory were available. But despite this promise, the 128K would have the shortest life of all the official Sinclair machines. It was too little too late. In April of that year, the Sinclair brand name and the rights to the Spectrum would be sold to Amstrad. They agreed to take on the 50,000 128Ks which were in the process of being assembled⁴³, but by the end of the year the Spectrum 128K would be gone, replaced that Christmas by the new Amstrad Sinclair Spectrum +2.

In Spain, Investrónica was replaced as official Sinclair distributor by Amstrad’s existing wholesale partner, Indescomp. Bitter reward for dragging the Spectrum forward where Sinclair would not and arguably preserving its relevance so that Sugar could work his cosumer product magic on it. Investrónica would go on to bring the Atari ST to Spain – an announcement made the previous year, alongside the reveal of the 128K, showing that they weren’t planning to keep all their eggs in the one basket – and to embrace the IBM PC compatible. But they were not quite finished with the Spectrum just yet.

The Inves Spectrum+

The Spectrum 128K was an official Sinclair machine, even if it was one seemingly forced upon its reluctant parent company. The Inves Spectrum+⁴⁴, despite wearing a case from the same official Sinclair moulds and a keyboard from the same supplier, is very much a clone. Released just before Christmas 1986⁴⁵, it was an attempt to retain a foot in the Spectrum market in Spain. There were still many invested users and a large body of compatible software, with a market rejuvenated by budget games especially⁴⁶. And it was, after all, a market which Investrónica had created. “Based on this fact, Investrónica retains the moral right to perpetuate the Spectrum Plus model, apart from Indescomp, Amstrad and even Sinclair”, declared Primitivo de Fransisco in his review for MicroHobby⁴⁷.

Unfortunately, for a machine sold on its implicit compatibility⁴⁸, this is where it fell down⁴⁹. Over the following months it would slide in the eyes of the magazines from fully-compatible to “halfway”⁵⁰ to having “serious incompatibility problems”⁵¹. It used the same Spanish version of the 48K BASIC ROM as was eventually made available for the Sinclair Spectrum+ and shipped with the Spectrum 128K. Alongside the localised messages and support for extra characters, this introduced a number of incompatibilities. Chief among these was its use of an area of memory which in the original Spectrum ROM was unused and filled with $FF. Many games used this space in setting up a custom interrupt handler, causing them to fail on the Inves.

The incompatibilities escalated at the hardware level. This was a Spectrum+ clone, designed to be low-cost. The inclusion of a Kempston-compatible joystick port was as sure a sign of the machine’s intended market as the external keypad of the Spectrum 128K. The machine’s improvements were missing here. Also absent were any official Sinclair components.

The ULA was replaced by a part designed by Flare Technology⁵² and manufactured by Texas Instruments using their TAHC10 process. This was a purely digital part with a smaller capacity than Ferranti’s gate array, meaning that some functionality – reading the keyboard and handling cassette input and output – was handled by additional discrete logic. This is a time when genuine ULAs were available for the repair market⁵³, and just a year or so before Didaktik would begin to manufacture the first run of Gamas containing Sinclair ULAs. The TAHC10 used timings which differed greatly from those of the Sinclair Spectrum – specifically around the point at which the vblank interrupt is triggered – meaning that timing-critical code did not run correctly.

Elsewhere, the hardware is a mixed bag. There is 64K of RAM, with the lower 16K inaccessibly behind the ROM⁵⁴, but there is no contention for access to it. The Z80 CPU runs at the slightly faster 3.54Mhz of the Spectrum 128K. There is the built-in joystick port. The keyboard is as good or bad as that of the Spectrum+ and 128K.

I’ll leave the final words on the machine to Miguel Ángel Rodríguez Jódar and César Hernández Bañó, whose excellent post-mortem of the Inves Spectrum+ is linked from the ‘Resources’ section below:

It is a machine that gives the impression that it has been made with a good idea on paper, but a hasty implementation that has not allowed time to test the machine. … Showing the black points of the Inves Spectrum+ “in the past”, with today’s technology, leaves the work of those engineers in a situation that, for me, is unfair.

Most Spectrum clones were developed in countries where ideas like intellectual property and copyright were strange, foreign concepts. Spain in 1987 – a newly-fledged member of the European Union – was not such a country. While the matter never reached court⁵⁵, its likely that lawyers’ letters were exchanged. Whatever their moral right, Investrónica’s infractions were blatant and untenable. The Inves Spectrum+ would disappear from sale and Investrónica would move on with the Atari ST, the NES, and its IBM PC clone business.

Resources

• MicroHobby magazine.

• ZX Magazine.

• The Inves Spectrum+, in depth.

• ZX Spectrum ROMs.

• Noel’s Retro Lab looking at the Inves Spectrum+ part one, part two.

• Fred Harris reviews the Spectrum 128K for BBC Micro Live.

• The Retro Shack with an impeccably-researched video on the 128K.

Notes

Timex

The company which would eventual become Timex Corporation began in 1854 as the Waterbury Clock Company¹. They specialised in producing domestic alternatives to the expensive European clocks and pocket watches of the time, manufacturing both for themselves and on behalf of others. They changed owners and name many times over the subsequent 100 years. The 1950s saw the company – now the United States Time Corporation – returning to their roots as a supplier of inexpensive and reliable timepieces. Using the techniques they had developed in the armaments industry, they combined simplified design, newly-developed materials, precision tooling, and automation to produce a reliable and cost-effective product. These skills and experience were in demand, which saw Timex simultaneously expand as a manufacturing sub-contractor to others. From 1945 they manufactured cameras for Polaroid. They expanded with factories across Europe and Asia in addition to the United States.

Timex Sinclair

The histories of Timex and Sinclair meet in the Scottish town of Dundee, where Timex assembled the ZX81 for Sinclair. By the start of the 1980s Timex’s consumer wristwatch business had been feeling the pressure from cheaper Far Eastern manufacturers for some time. Computers seemed like a natural area for them to expand into. From Sinclair’s point of view, having a partner to shoulder the costs associated with breaking the American market, where they already offered limited mail-order sales, offered nothing but advantages. It would be a brief and ill-fated partnership.

Timex Sinclair’s first product – the TS1000², a slightly-modified ZX81 with NTSC video support and 2K of RAM – was announced³ on April 20th 1982. Three days later, Clive Sinclair introduced the ZX Spectrum to the world. For a watch company, Timex had pretty lousy timing. A cynic might also wonder how this was allowed to happen. Surely someone on the Timex side must have been aware that Sinclair’s announcement was imminent, if only because the Dundee factory had been preparing for production of the new machine⁴.

TS1000

The TS1000 was released in July 1982. Aided by the low price, an extensive advertising campaign, and generally positive reviews⁵, demand was high. Timex Sinclair would lean heavily on the ‘first computer under $100’ shtick. It wasn’t a lot of computer, but it wasn’t a lot of money, either. The issue of Byte which reviewed the TS1000 listed it for sale at $89, alongside the Atari 400 for $249 and the VIC-20 for $189. 600,000 TS1000s had been sold by the end of that year⁶ – a claimed 30% of all home micro sales – placing it joint second with the Apple ][ behind the VIC-20’s 750,000. 70,000 of these sales came via an agreement with American Express⁷. The TS1000 was inducted into the Smithsonian Institute’s permanent collection⁸. By whatever criteria you wanted to measure it, the TS1000 was a success.

Many business textbooks have been written on the subject of following-up a successful product. There is universal agreement that what you don’t do is tease a bigger and better replacement before you have it ready to go. At the January 1983 CES in Las Vegas, Timex Sinclair announced the TS2000 series⁹, widely dubbed ‘the American Spectrum’. They would have much improved graphics and sound. A cartridge slot would alleviate the woes of loading software from tape. Bank-switching would allow for up to 16Mb of memory. There would be the 48K TS2048 for $199.95 and the 16K TS2016 for $149.95. It had been an almost-certainty since its launch in the UK that Timex Sinclair would release their own version of the Spectrum, but the way the announcement was handled seemed poorly judged. Sales of the TS1000 instantly declined. By March 1983 some retailers were offering it for a fire-sale $49.95.

To muddy the waters further, rumours of another Timex Sinclair machine soon began to circulate. “Confusion seems to be the order of the day at Timex,” one editorial observed¹⁰. The TS1500¹¹ was said to be a TS1000 with 16K of RAM. “We haven’t announced any product like that,” said Dan Ross, Timex Sinclair’s Chief Operating Officer, in an interview with Timex Sinclair User¹². The same issue carried its officially announcement¹³.

The TS1500 went on sale in June 1983. Enclosed in a case almost identical to the Spectrum’s, that would be the only thing it borrowed from Sinclair’s more recent machine. Inside the TS1500 was basically a cost-reduced version of its predecessor. The Ferranti ULA had been replaced by one manufactured by NCR, and as predicted RAM had been increased to 16K, but otherwise it was the same silent, monochrome machine. Even as the first 16K computer for under $80 it was met by general bemusement.

TS2068

The TS2000 series would be released in November 1983 in the form of a single 48K model, the TS2068¹⁴. To write it off as merely ‘the American Spectrum’ would be a mistake¹⁵. While not exactly revolutionary, the TS2068 took a number of evolutionary steps, some of which it would take the official Spectrum years to catch up with – and some which it never would in its lifetime – and all just a year and a half after the Spectrum’s launch. The TS2068 included an AY-3-8912¹⁶ sound chip; the Spectrum wouldn’t get the same until the Spectrum 128K two years later. There were two joystick ports, one on either side, read via the sound chip and not following either the Kempston or any of the key-base schemes. The keyboard was an improvement, if for no other reason than it had a full-size space bar.

Externally, besides the keyboard, the most obvious change was the cartridge port, concealed under a flap to the right. It is said to have been added due to the machine’s designers’ frustration with loading software from cassette¹⁷. Each cartridge can hold up to 64K as eight 8K banks which can be mapped individually into the Z80’s address space.

On the rear of the computer, from right to left, are RF out – there’s a channel selection switch on the underside of the case – power, cassette in and out, composite video out, and an expansion connector behind a cover. The expansion connector is not compatible with the one on the Spectrum.

Inside, the heart of the machine is a SCLD (Standard Cell Logic Device¹⁸), the TS2068’s version of the ULA, performing all the same functions as its Spectrum counterpart, along with managing the additional bank-switching features. Contemporary reviews credit the chip to NCR (National Cash Register) Semiconductor, although later machines would contain chips bearing the Ferranti Portugal mark.

Turning the TS2068¹⁹ on – without a cartridge installed – we are greeted by a pair of copyright messages, one for Sinclair and the other Timex.

What should have been the biggest new feature of the TS2068 over the Spectrum was the introduction of the ‘enhanced display’ modes. The regular 256 by 192 pixel, 32 by 24 character colour screen was supplemented by three new variants. All used an extra 6912 bytes of memory. The first simply added a second identical screen, which would allow double-buffering – drawing into the ‘off screen’ memory while the ‘on screen’ was being output to the monitor, a technique used by games to reduce the flickering which could occur when attempting to update display memory while it was being drawn. The other two were far more interesting.

‘Extended color mode’ used the second bank of display memory to hold an attribute byte for each byte of pixel data in the first, meaning separate colour attributes could be specified per 8x1 pixel cell, rather than the standard 8x8. Decades later, clever hacks like the BIFROST* Engine²⁰ would manage to replicate something similar with the original Spectrum hardware, albeit only in a sub-area of the screen and at a major cost in terms of performance.

‘Full width mode’ produced a 512 pixel, 64-character-wide display. Pixel data for alternate columns was drawn from the first and second display memories in turn²¹ and output by the video system at twice the pace of the other modes. “64 characters, plus two 8 character-wide margins exactly fills a standard 80 column format page for word processing applications,” says the manual²², in an attempt to persuade us we didn’t really need 80 columns. Along with the improved clarity of the composite output this would have been a boon to professional software, if only the number of programs which used it had been larger. Among the number which did not was the TS2068’s own in-ROM BASIC. It would take a heavily-patched version, in the form of Zebra’s OS-64²³, to fix this issue.

The state of the TS2068’s ROM in general was indicative of a development process which ran out of time, despite pulling in such big guns as The Complete Spectrum ROM Disassembly author Ian Logan²⁴ to help. There were the kernels of some good ideas, innovating on the Spectrum ROM on which it was based, like the ‘function dispatched’ which provided a stable API for calling in-ROM BASIC routines from other languages. But for every improvement there was a regression. No longer able to fit within 16K, bank-switching – still heralded by the user manual as giving access to 16Mb of memory – was used to juggle between the main BASIC ROM and the 8K Extension ROM, mapped into the start of memory. Marketing would total these, and along with the 48K RAM would declare the TS2068 a 72K system.

But by far the TS2068’s ROM’s biggest short-coming was its lack of full compatibility with its British cousin. This would soon come in the form of build-it-yourself projects which would allow for the installation of the ROM from the 48K Spectrum, either internally or via the cartridge slot, some purportedly based on Timex’s own planned “Chameleon” Spectrum emulator cartridge²⁵. “All the software that we envisioned when we originally purchased our 2068s is available NOW if you have the Spectrum ROM,” wrote one reviewer²⁶. Similar hardware fixes for the incompatible expansion connector were equally quick to appear.

Ultimately, whatever its advantages or disadvantages, the TS2068 never really got a chance to prove itself in the market. In early 1984, barely four months after its release, Timex would bow out of the computer market altogether. The world never got to meet the TS3068²⁷. The partnership with Sinclair was ended, although it’s probably reasonable to assume that it had been strained for some time. No matter how apathetic Sir Clive may have been to the Spectrum by this point, Timex’s announced intention to ‘improve’ it for the American market must have rankled – no matter that it became necessary after Sinclair’s own NTSC version of the ULA failed FCC testing. Sinclair is known to have declared that Timex has “fouled up” the Spectrum²⁸. There was probably also some small amount of schadenfreude that it was the time and cost of this re-engineering which would ultimately majorly contribute to Timex Sinclair’s downfall. Which is a shame, since the TS2068 contained the germ of many good ideas, which if given a little extra time to mature could have formed the basis for at least a Spectrum v1.5. Unfortunately, the only thing Sinclair seemed to take from their partnership with Timex was the TS1500’s innovation of re-launching the same old computer in a shiny new case.

But this would not be the end of the road for the TS2068. The machine would live on in the hearts and minds of its users, who through magazines and user groups would keep its spirit alive, and chronicle the trickle of new software and hardware released for it. More practically, the TS2068 would continue to be produced on the other side of the Atlantic.

Timex Computers

The history of Timex Computers – as Timex’s Portuguese subsidiary would come to be best known – is a story for a different article. Suffice to say that it was chosen to manufacture the TS1500 and TS2068 for sale in American, and continued to do so after Timex Sinclair’s withdrawal from the market. Portugal was the one Western European market which the deal with Sinclair has ceded to Timex. Renamed the TC2068, the TS2068 would continue to be sold here in two forms. The first came in a black case, bundled with the TimeWord²⁹ word processor, which made full use of the 64 column ‘full width’ display mode. The other – the “Silver Avenger”³⁰ – apparently under no misapprehension about its intended use, came bundled with a Spectrum ROM cartridge. Both models featured redesigned main boards with fully Spectrum-compatible extension connectors³¹.

The Portuguese market wouldn’t be the end for the TC2068, either. It was sold into South³² and Central America³³, back into the US in a small way³⁴, and into Poland. The Unipolbrity 2086³⁵ would eventually be assembled in Gdańsk from components imported from Portugal, combined with a better keyboard³⁶. These too were accompanied by the Spectrum ROM emulation cartridge, and while it’s good to know that the TS2068 lived on, at least for a little while, it’s sad that it was mainly while pretending to be a lesser machine.

The TS2068 is a story of missed opportunities. In a parallel universe, the Spectrum+ contained a sound chip and a couple of enhanced graphics modes inside its shiny new case.

Resources

• timexsinclair.com is an amazing treasure house of material.

• Sinclair Timex Users Magazine

• Timex Sinclair 2068 Technical Manual

• Time Designs Magazine

• Fuse

Notes

In a video review of the pseudo-retro 8-bit AgonLight computer¹, Noel of Noel’s Retro Lab² posed the question, “What defines the character of a computer?”. He concluded that it was the video display. It’s an opinion I fully agree with. Most 8-bit systems in particular are so simple and similar – either a 6502 or Z80 CPU with some RAM and ROM – that the only feature distinguishing between them is the video they output.

One of the areas of computing history we’re exploring on this blog is the world of Spectrum clones, and the Spectrum is one of the very best examples of a computer’s display defining its character. There is no mistaking a Spectrum screenshot, especially one which shows off its full multi-colour attribute-clash glory.

One of the most interesting aspects of Spectrum clones is the inventive ways in which that iconic display output is reproduced. The genuine Spectrum uses a semi-custom chip – the ULA – to integrate the circuitry which generates the video signal. Most Spectrum clones came from Cold War era Eastern Europe, where comparable technology was not readily available to the clone builders. Thus they resorted to replicating the functionality excapsulated in the ULA using individual logic chips. To appreciate exactly what an achievement this was, I think it’s worth taking the time to examine what the main characteristics of the Spectrum’s video output are and how it is produced. In so doing we can appreciate the genius of its inventor, Richard Altwasser³.

The Spectrum has a 256 by 192 pixel bitmap screen. While this is nominally the same as its predecessor the ZX81, the approach taken and the effects achieved are radically different. The ZX81’s screen is character-mapped⁴, limited to drawing only a predefined set of glyphs⁵. It uses a variable-sized area of RAM – the DFILE – to store 25 lines of up to 32 characters. When it is time to draw these to the screen, the bitmap representation of these characters are looked-up in ROM and passed to a shifter, which outputs them pixel-by-pixel. All of this is done by the Z80 CPU, taking up approximately 80% of the available processing time each video frame. In contrast, the Spectrum allocates 6144 bytes of RAM as permanent display memory, able to display any kind of bitmap graphics. Responsibility for outputting this data to the video display was taken away from the Z80 and given to custom circuitry, allowing the CPU to dedicate 100%⁶ of its time to running programme code.

But what would give the Spectrum its distinct visual style – and indeed earn it the ‘Spectrum’ name – was the addition of colour. To achieve this, each 8x8 pixel cell on screen was assigned an attribute byte, containing independent foreground and background colours. These provided a memory-efficient means of adding colour, using just an extra 768 bytes of RAM, stored directly after the bitmap data. The approach was innovative, earning Altwasser a patent on the technique⁷⁸. It is responsible for the iconic colour clash, and for a few other eccentricities of the video circuitry which any Spectrum clone would have to recreate.

The following description of how the Spectrum’s video signal is generated is taken wholesale from Chris Smith’s book How to Design a Microcomputer⁹. Any errors introduced while over-simplifying it are entirely mine.

Generating a video signal is, at heart, a matter of counting. The PAL¹⁰ signal which the Spectrum produces¹¹ consists of 50 312-line frames every second. The video subsystem is driven by a 7Mhz clock, ticking 448 times per line. Both the vertical line count and the horizontal column count use 9-bit counters.

For simplicity, the vertical counter starts at the point the first line of the Spectrum’s screen is output and counts the 192 lines. It then counts 56 lines for the bottom border, the equivalent of 12 lines for the vertical blank while the electron beam is travelling back to the top of the screen, and finally another 56 lines for the top border, before resetting the count to 0 and starting again. Likewise, the horizontal counter begins with the first column of the Spectrum’s 256-column screen, followed by 64 columns for the right border, the equivalent of 96 columns while the electron beam moves back to the left of the screen, and finally 32 columns for the left border. This means that while the vertical counter is in the range 0-191 and the horizontal counter is in the range 0-255 they together give the screen coordinate of the pixel which is being output.

Of course it isn’t quite as simple as that. Firstly, pixels are stored in memory as the bits within a byte. Before they can be shifted one-by-one into the video output, the byte containing them needs to be fetched. Secondly, the monochrome pixel data needs to be combined with the colour information stored in the attribute data. Since attribute data has a horizontal resolution of the same 8 pixels, we can see that before any run of 8 pixels can be output, the video subsystem needs to fetch two bytes from display memory: one containing the pixels and the other containing the colour attributes. From this stems two further idiosyncrasies of the Spectrum’s design. The first is contention, the possibility that both the video subsystem and the CPU will need to access the display memory at the same time. This is arbitrated at the level of the entire lower 16K block of RAM containing the display memory. In these cases the video subsystem takes priority, removing the clock signal from the CPU while it retrieves the data it needs to keep the video signal flowing.

The second idiosyncrasy should be familiar to anyone who has sat and waited while a Spectrum game loads from tape, although they may not have be aware of what exactly they were seeing. Anyone who has ever tried to program bitmap graphics on the Spectrum, however, will be painfully aware of what is going on.

Here we are watching data being loaded from tape straight into the Spectrum’s display memory. First the bitmap data, and then finally the attributes. The data is being loaded linearly but the lines do not appear on the screen in the order one would naively expect. The reason for this is that the Spectrum’s display memory is not laid out in a strictly linear fashion. And the reason for that is efficiency.

The diagram below¹² shows how the horizontal and vertical counters maintained by the video subsystem – and thus the coordinates of each pixel – relate to the address of the corresponding bitmap data in RAM. The display memory begins at $4000 in the Spectrum’s memory map, but from the point of view of the video subsystem, which can only see the lower 16K, it begins at $0.

The horizontal component of the address is easily comprehended. The lower 3 bits – H2-H0 – are ignored since we are dealing with byte memory addresses, each containing 8 pixels. H8 is ignored because once the counter passes 256 we are past the screen and into the borders. The remaining 5 bits – H7-H3 – give us 32 bytes per screen line.

The vertical component is more involved. It helps to imagine the screen as divided into 3 equal parts, with each part consisting of 64 lines or 8 character rows, where each character is 8 lines tall. Thus bits V5-V3 index a character row within a screen part and V2-V0 index a line within a character row. But why are they arranged as they are in the address, with the row index before the line? One benefit of this layout is that in order to increment the address pointer to the next line – as one might if you were writing a glyph into a character cell – you need only increment the high byte of the address, an efficient operation on the Z80. But this is likely just a happy side-effect of Altwasser’s real intent.

To understand what is being optimised here, it’s necessary to know a little about DRAM memory chips. The lower 16K of RAM containing the display memory in the Spectrum was made up of eight 6114 chips. Each had a 14-bit – 16K – address space and was responsible for maintaining 1 bit of information. Due to both their internal architecture and the limitations of their packaging, the 14-bit address space was divided into a 7-bit row and 7-bit column address. To select a particular bit to read or write, the system puts the row address on to the chip’s 7 address pins, followed by the column address. This setup takes a certain amount of time. To aid efficiency, DRAM also supported ‘page mode’¹³, where the row address would be held constant while the column was changed, speeding the read/write setup time.

The Spectrum’s video generation circuitry makes use of DRAM page mode to make fetching the byte of pixel data and its associated attribute byte as efficient as possible. How it achieves this is typically inventive. The orthodox setup would see the row representing the high part of an address, with the column the low part. Thus page mode would allow faster access to data which is close together, such as the bytes of program code. In contrast, the Spectrum’s video subsystem uses the row to represent the low part of the address. Due to the layout of the display memory, this is common between a byte of pixel data and its character cell’s attribute byte.

The diagram below shows how the attribute address is derived from the horizontal and vertical counters. Comparing it to the pixel data address we have already met we can see that they share the lower 8 bits, which allows for the page mode access efficiency at the cost of the slightly unusual memory layout.

So why the Spectrum?

The above describes in very rough terms the operation of the Spectrum’s video subsystem. A third party, if they wished to produce a clone which remained compatible with the original, would need to analyse, identify, and replicate this behaviour. And they did. Many separate individuals and groups managed to unlock Altwasser’s secrets and produce their own version, usually assembled from discrete logic chips.

But there remains the question: why? Particularly in the context of Eastern Bloc countries, where most of the cloning happened. Why the Spectrum? Why go to such lengths to copy this particular machine? Nostalgia aside, the Spectrum was far from the best computer of its time. Its killer feature was its low price, which stemmed from its simplicity. It was a Z80, some ROM and RAM, and a semi-custom chip containing the eccentric video subsystem. By way of comparison, a Z80 and memory got you most of the way to a CP/M system, where a custom BIOS¹⁴ would allow you to use whatever simple display system you had access to¹⁵.

A mark in favour of the Spectrum was the large library of software available for it. But the vast bulk of that – particularly in the later part of its lifetime – was games¹⁶. While there’s certainly plenty of evidence now that being introduced to the world of computing via games is an excellent gateway to a career in ‘more serious’ computing¹⁷, this was not a given at the time, and would surely have been hard to sell to the kind of authoritarian bureaucrat whose permission would have been required to dedicate resources to the project of cloning the Spectrum. There was an equally large library of software available for CP/M, most of it of a more serious bent and thus more likely to meet with official favour.

I’m sure that as we explore the world of Spectrum clones and the histories of the people who built them we’ll get many different answers to this question. The cleverness of the engineering is certain to play some role. Because ultimately, the genius of Altwasser’s design wasn’t that it was so simple it could be copied and reimplemented by those with access to only more basic technology, it was that it was so efficient, so elegant, and so successful, that they would want to.

Notes

The heart of the ZX Spectrum is the ULA, the semi-custom microchip which, among other input-output duties, generates the distinctive display. It is what makes a Spectrum a Spectrum. Most Eastern Bloc Spectrum clones replicate its behaviour with scores of discrete logic chips, with varying levels of technical ingenuity and compatibility. The Czechoslovakian Didaktik company took a more direct route. The Gama was unique in its use of a genuine Spectrum ULA. These were still being produced into the late 1980s to supply the after-market repair industry. The last rolled off the Plessey¹ production line in 1989². This left Didaktik with a dilemma: How do they continue production of their Spectrum-compatible computers if they can no longer get the most important component?

A solution would present itself in the shape of Angstrem³, the Soviet semiconductor manufacturer, based in Zelenograd, Moscow. Boris Malashevich, historian of Soviet computing, former Chief Specialist at Angstrem, and “uncle of the domestic microprocessor”⁴, recounts the story in his article Zelenograd Household and School Computers⁵, published in 2008:

In 1990, the head of the microprocessor department of [Angstrem’s sister Research Institute of Fine Technology] NIITT P.R. Mashevich was on a business trip in Slovakia⁶ and saw in the store a Sinclair-like Didaktik Gama computer – a Slovak clone of the ZX Spectrum. At his request, the head of the enterprise where he was … organized a visit to the Didaktik Skalica company. Having familiarized himself with the computer, P.R. Mashevich proposed replacing the medium-integration IC used in it with one semi-custom LSI. Didaktik management met the proposal without enthusiasm, doubting its feasibility, but agreed to try LSI. For the work, Mashevich was given an electrical circuit and a sample Didaktik Gama.

…

Samples of LSI, along with relevant documentation and recommendations for use, were transferred through the embassy to Slovakia … There was no answer for a long time. Later it turned out that the LSI was handed over to the developer [at] Didaktik Gama for testing, but he did not understand it in detail, and after a careless and unsuccessful attempt to use it, he declared the LSI unsuitable for use.

To clarify the situation, P.R. Mashevich and B.V. Ilyichev went to Slovakia, taking with them several samples of their boards. When they were put into computers, they immediately started working without any differences from Slovak machines. As a result, Didaktik Skalica for a number of years became a stable consumer of products from the Angstrem plant, and not only the … controller, but the entire LSI set that grew around it.

So, a heart-warming tale of Soviet engineers helping out their backwards Czechoslovakian comrades with a complex technical problem. But there is an alternative version⁷ of this creation myth, one which lacks an official publication but is simpler and contains fewer embellishments⁸:

Victor Klemon, an extremely interesting person and a capable organizer, contacted Soviet developers in 1989 … [who] agreed to supply the kits (ULA + RAM + ROM) for [Didaktik’s] production. Of course, in Zelenograd they produced not an analogue of the proprietary ULA … but their own development based on the BMK … [the T34VG1].

Whichever version contains the most truthiness, the relationship bore fruit in the form of the T34VG1⁹ (Cyrillic: Т34ВГ1). This was an application of Angstrem’s Base Matrix Crystal¹⁰ (BMK) (Russian: Базовый матричный кристалл - БМК) technology, a gate array similar to Ferranti’s Uncommitted Logic Array. The T34VG1, like the ULA it was designed to replace, handled video signal generation, clock generation, keyboard reading, tape signal input and output, and beeper audio output. It also had an additional input pin which could be used for implementing extra interfaces such as a Kempston joystick¹¹.

The first micro to contain the T34VG1 would be the Didaktik M, launched in 1990.

Taking a Look

This example is complete with the original box.

The M is presented in a hard plastic case, similar to the Gama, although longer and slimmer. Inside is the computer and its power supply. The manuals which would originally have been included are not here. I’m trying to track down versions online.

The design of the M’s case is by Marián Višňovský¹². In both appearance and manufacture it is light years ahead of most other Eastern Bloc Spectrum clones, and in my opinion is up there with the best-looking of the 8- and 16-bit micros. The keyboard isn’t completely terrible, either. The chunky RESET button requires CAPS SHIFT to be held at the same time to trigger.

On the rear of the case, reading right to left, we have RF and composite outputs; power and cassette (‘mangnetofon’) in; the expansion edge connector, which is almost-but-not-quite compatible with the Spectrum; and edge connectors for Kempston and Sinclair joysticks. The former makes use of the additional function of the T34VG1, while the latter connects via the data lines used by the keyboard matrix.

On the underside of the machine is a paper sticker with a 9012 manufacturing code. I was also surprised to find an intact warranty seal over one of the eight screw holes.

Once unscrewed, the top of the case lifts away, leaving the keyboard in place above the main board, which is another nice indication of the care put into the design.

Moving the keyboard aside, one of the first things to strike you is how uncluttered the motherboard is. In part this is due to the simplification brought by the T34 – here labelled ‘ULA1’ – but also because the M offers fewer extras compared to the Gama. Gone is the 8522 peripheral interface. RAM is also reduced from 80K to just 48K. The eight Soviet “KR565RU5G” (Cyrillic: “КР565РУ5Г”) RAM chips on the right of the board provide 64K, the lower 16K of which is inaccessible behind the ROM.¹³ Both the RAM chips and the T34 have 9010 date codes.

The CPU is a Zilog Z80, here clocked to 4Mhz, up from the Spectrum’s 3.5Mhz. It has a 9037 date code. The ROM is Microchip branded. There are a few extra Tesla LS logic chips dotted around the board. Video output is managed by a combination of LM1886N – to convert the T34’s digital output to analogue – and LM1889N.

One final observation: the black blob of the T34 is, to me at least, superficially reminiscent of the epoxy-covered NES-on-a-chip found in Famiclones¹⁴.

Turning it On

Powering the M on, we see the copyright message “© 1990 DIDAKTIK M BASIC”.

Playing around with the border colour reveals something interesting.

No, don’t adjust your web browser, your eyes do not deceive you, the Spectrum’s 256x192 screen is being drawn as a square.

The eccentricities don’t end there. On paper, the 4Mhz CPU sounds great compared to the Spectrum’s 3.5Mhz. But the T34VG1 was designed in such a way that all memory access by the CPU now face wait times¹⁵. In the Spectrum, only the lower 16K containing the screen memory was subject to contention when the ULA would halt it if it attempted simultaneous memory access. Malashevich seems to treat this regression as an innovation:

… video memory, separate and with its own controller in English and Slovak computers, was circuit-technically combined with common RAM with a single controller. The RAM used dynamic memory, requiring periodic regeneration of information. This process is similar to the process of image regeneration on the screen, and they were combined in one device.

The faster processor speed in the M was meant to compensate for this slowdown, and in one respect it does. Running Noel’s BASIC benchmark¹⁶ gives a time in the same 87 second region as a genuine 48K Spectrum. But at the same time the faster clock speed introduces incompatibilities with time-sensitive code, such as that which counts instructions and updates the border colours as the screen was drawn.

T34

The designation ‘T34’ was given to the chipset by Angstrem engineer Yu.L. Otrokhov, who served as a tank driver in his youth¹⁷. Otrokhov was the chief designer of the T34VM1 (Cyrillic: Т34ВМ1), a from-scratch reimplementation of the Z80 which Angstrem worked on at the same time as the T34VG1, and which when finished just happened to feature the same bugs and errata as the East German U880¹⁸.

As with the rest of this history, there isn’t a clear picture of what else Angstrem was working on in terms of microcomputers at this time. Some¹⁹ claim that the ‘Zelenograd engineers’ were already working on a BMK-based ULA analogue based on the Baltik Spectrum clone, and that it was this already-completed product which Didaktik saw and chose to use. This would provide a 3rd version of the T34VG1’s creation story. However it happened, by the end of development Angstrem had clones of the ULA and the Z80. With the addition of RAM and ROM²⁰ that was everything you needed to build a computer.

Let’s take a moment to appreciate the accidental aptness of the ‘T34’ name. Designed during the later half of the 1930s, the Soviet T34 tank²¹, while it contained a few innovations, was not cutting-edge technology. It was designed with the available production capabilities firmly in mind, with cost and speed of production leading factors. The undoubted skill and courage of their crews aside, the T34’s strength lay in sheer weight of numbers. Over 80,000 were built during the course of the Second World War.

Returning to the start of the 1990s, Angstrem’s next move was just good business: they release the T34 chipset (including the ROMs, complete with Didaktik copyright message) onto the rapidly-opening market. As the Soviet Union collapsed, if you found yourself in charge of the necessary facilities with an under-employed staff skilled in electronics assembly, entrance to the home computer business was only an order away. What followed was a Cambrian Explosion of Speciclones, as factories from the Baltic to the Black Sea, from Brest to Izhevska, churned-out their own variants.

These machines were much of a muchness, as you would expect from their common components. They varied only in the cheapness of their cases and the awfulness of their keyboards. If we single any out for special mention it should be the ANBELO/C. This is the machine which Angstrem were working on as the prototype of the T34 machines, either before contact with Didaktik, or while waiting for Didaktik’s approval of the newly-delivered chipset, depending on which story we believe. The instructions which accompanied the completed machine contain the following:

The purpose of this development was to create a new computer element base, which is not only software, but also hardware compatible with computers of this type produced in European countries. The household computer “DIDAKTIK 2”, produced since 1990 and already found in our country, was chosen as the prototype.

Also worth a mention is the ALF (or Elf) (Russian: “Эльф”) TV Game²². Packaged like a game console, it sidesteps the keyboard problem by having only control pads. Games were delivered via ROM cartridge²³. An attempt to counter the influx of Famiclones, the obvious drawback was that Spectrum games had to be adapted to work from cartridge, at a stroke destroying the advantages of availability and price which came with easily-copiable cassette-based games.

Production of the T34 chipset and micros based on it would continue for a few short years, until ultimately a combination of PC-compatibles for serious work and Famiclones – such as the near-ubiquitous Dendy²⁴ – for cheap (and reliable) fun would come to dominate.

Didaktik would release one more home computer using the T34VG1. The Kompact included a 720K 3.5” disk drive. It looked lovely but was too little too late.

Additional Resources

• Schematics for the Didaktik M (plus many other machines) https://z00m.speccy.cz/?file=hardware.

• The ‘Clueless Engineer’²⁵ video series on the Didaktik M on YouTube.

Notes

The Didaktik Production Cooperative was formed in Czechoslovakia in 1971¹. It was based in the town of Skalica, just on the Slovak side of the Czech/Slovak federal border. As their name suggests, Didaktik’s initial focus was education. Their Alfa and Beta micros were used by many schools in the region. These were enhanced clones of the PMD 85 micro developed by Tesla.

The Czechoslovakian Tesla company was founded in 1946 as an amalgamation of nationalised engineering firms, among them the local branches of foreign companies including Philips and Siemens. The name was originally chosen to honour American-Serbian inventor Nikola Tesla. Representatives of the Yugoslav government were present at the foundation ceremony. A few years later, an ideological falling-out lead to the name being hastily retconned. It now meant “TEchnika SLAboproudá”, “low-current technology”.

The many Tesla branches were responsible for the design and manufacture of all types of electronics within Czechoslovakia, both military and consumer, including televisions and radios. These in turn used domestically-produced components such as transistors and eventually integrated circuits. Tesla produced version of the standard 74-series logic chips and even a clone of the Intel 8080 CPU. This would be used in the PMD micros, designed and built by Tesla Piešťany².

Didaktik would seek inspiration for their next machine – to be called the Gama, naturally – from further afield.

The Spectrum had become the most popular home micro in Czechoslovakia for many of the same reasons it had succeeded elsewhere in Europe. It was relatively cheap, readily available, and had plenty of attractive software. As Amatérské Radio³ wrote:

The main factors in establishing home computers are their cost and software availability. There is plenty of awareness of these two requirements overseas and thus foreign companies put a lot of effort to address these matters. This is reflected in this country by the fact that the most popular home computer makes are the various Sinclair company models.

The rubber-keyed 48K’s small size also undoubtedly played a role in its popularity. While you could find Spectrums for sale in state-run shops such as the foreign currency Tuzex⁴ stores, it was more likely to have been brought into the country by deep-pocketed workers returning from the West, either for family or for sale on the black market.

Nevertheless, the machines remained hard to find and out of the reach of most of the population. Meanwhile, domestic computer magazines such as Mikrobaze carried articles about the Spectrum, along with BASIC listings and the schematics for hardware add-ons, driving interest and demand.

The Gama was designed and put into production in a matter of months in late 1987. It’s release was accompanied by a television advertising campaign⁵.

Although cheaper than an imported Spectrum, the price was still relatively high, and availability was limited. Mikrobaze⁶ carries a contemporary account of the tribulations facing potential buyers:

When a shipment of 50 Didaktik Gama computers was supposed to be delivered to the store in Spalená Street in Prague at the pre-Christmas market, a queue began to form in front of the store as early as 6 o’clock in the morning. At 9 o’clock there were about 35 eager citizens standing in line. The last … entered at about 12:30 pm.

For security reasons, only 15 people could wait inside the store. As always with extraordinary events, the power of human solidarity was shown here - those waiting in the store argued with those outside. It was above zero, but an unpleasant wind was blowing.

At 4:10 pm they finally brought the desired computers. The sale started at 4:20. It is unfortunate that not everyone could be served that day, because it takes experienced sellers about 10 minutes to demonstrate and try out one Didaktik.

Not sure how I’d feel if my chance to buy the computer was missed because the previous chap’s trial game of Hungry Horace took too long.

Taking a look

But if you were one of the lucky ones, this is what you would have got.

Included are two booklets, the “Zárucny List” (“Warranty Letter”) and the “Pripojenie Periférií” (“Connection of Peripherals”) manual. You can find scans of these on archive.org here, and an English translations in the Archive.

This is the ‘89’ revision of the machine, the 3rd produced. The original ‘87’ and updated ‘88’ versions came in black cases. It’s a shame none were produced in blue or orange cases like the Alfa and Beta.

The keyboard is probably on a par with that of the Spectrum+, and of course the keys carry all the usual BASIC keywords. The red LED is marked “bank”. Below it is the recessed reset button. From the amount of pencil on and around it, it’s seen some use.

At the back of the machine, starting from the right, we have RF and composite video out sockets. The 5-pin DIN carries 5v and 12v DC, along with cassette data in and out. The cable from the power supply has a second tail which carries the cassette lines to a second DIN. The expansion connector is compatible with the Spectrum (less a couple of voltage lines) and benefits from being an actual edge connector, as opposed to the non-standard connectors present on many other Eastern Bloc Spectrum clones. Finally we have a keyed 30-contact FRB connector which acts as a parallel port.

Opening the Gama up, the first thing you notice are the lovely turquoise chip sockets. The CPU is a ST Micro Z80A with a 1990 date code. There are images elsewhere online⁷ which show the machine using the East German U880, usually in its distinct white ceramic package. The choice of which CPU to use must have come down to what stock was available for that production run.

Along the bottom of the board is 16K of DRAM as 8 Fujitsu 8118 chips with 83 date codes, and above them another 64K as 8 Soviet “KR565RU5G” (Cyrillic: “КР565РУ5Г”⁸) chips with 88 date codes. The Gama has a total of 80K, with the upper 32K swappable between two banks, as indicated by the red LED next to the keyboard. The Motorola and Tesla logic chips in the bottom right corner manage the banking. It’s worth noting that this board contains a mix of metric (Soviet) 2.5mm and imperial (Western) 2.54mm pitch components.

The Tesla MHB8255A at the top left is a clone of the Intel 8255 Programmable Peripheral Interface chip. One port is used to control the bank switching, while the others are exposed through the smaller FRB edge connector.

To the top left is the shielded RF modulator. A LM1889N handles encoding the video signal.

But what glues all these components together and generated the video signal? Hidden in the bottom left corner, underneath the keyboard connection wires, we have…

…a genuine Ferranti-manufactured Spectrum ULA. This is the 6C001E-7 variant with an early 1987 date code.

The ULA

The use of a genuine ULA by Didaktik is, as far as I’m aware, unique among Eastern Bloc Spectrum clones. While much of the interest in these machines comes from the innovative technical ways in which their designers replicated the behaviour of the ULA, the presence of the genuine article is interesting in it’s own right.

Chris Smith’s excellent How to Design a Microcomputer⁹, the Bible for all things ULA, tells us that the E-7 variant was first produced for the Issue 4A Spectrum motherboards in 1984. It would be the final revision, with all previous quirks corrected, used in all subsequent 48K machines. But by the time the first Gamas were produced in late 1987, the Sinclair brand had passed on to Amstrad. Production of previous Sinclair machines had ceased in favour of the new +2, which used the 48-pin ULA from the short-lived Spectrum 128K¹⁰.

At the same time, there was still a lot of new old stock available. Massive over-production for Christmas 1985 had lead to the financial troubles which ultimately saw Sinclair selling out to Amstrad. Those machines still existed, unsold, and would turn up in unexpected places, such as as this Spectrum+ with the serial numbers filed off sold by the Yugoslavian computer company Iskra Delta¹¹ and available briefly in Czechoslovakia.

There was still a demand for the older ULAs. There were tens of thousands of 48K Spectrums in homes across Europe, many of them in need of repair. In addition to reclaimed components and old stock, the E-7 ULAs continued to be manufactured to meet this demand.

The “ULA Gallery” maintained by Byte Delight includes examples with date codes up to late 1989, by which time Ferranti’s semiconductor business had passed on to Plessey. The 1987 version we have here is therefore not unusual. It’s presence in a non-Sinclair machine sold in a foreign market is.

In his introduction to the “Connection of Peripherals” manual we saw earlier, Didaktik’s lead engineer Viktor Klemon writes:

At the end of 1987, after a short but intensive production preparation, we commissioned the production of the Personal microcomputer Didaktik Gama. By the end of 1988, we produced 8,000 units and we expect a significant increase in production - up to 25,000 units in 1989.

I have no evidence of whether or not the later target was met¹², but even if only the 8000 unit figure is correct, that is a lot of ULAs to source from somewhere¹³. It would be interesting to find out whether these were bought on the open market, or whether there was a deal of some type struck between Didaktik and either Amstrad or Sinclair Research to provide the ULAs. That would make the Gama at least a semi-official clone.

The most interesting – but also most unlikely – lead I’ve found in this direction¹⁴ introduces disgraced tycoon Robert Maxwell to the mix. Maxwell was originally from Czechoslovakia, and during 1985 was considered the front-runner to purchase Sinclair¹⁵, so there is a strong connection, but what his role was in the deal, if indeed there was one, is unclear.

Trying it out

Connecting the composite cable and powering the machine on, we see:

The altered copyright message and the chunkier font belie what is a lightly modified version of the Sinclair 48K ROM.

Unfortunately you cannot ignore the muddy quality of the image, or the fact that it is black and white. A cursory Google and a bit of prodding-around with a multimeter suggests that the signals coming out of the LM1889N are on the weak side. I’ll try to fix them in a future post.

Also in a future post we’ll revisit Didaktik and see how they were to inadvertently have a lasting effect on the world of Spectrum clones.

Additional Resources

• Various resources for the Gama (and other Spectrums) including schematics: https://z00m.speccy.cz/?file=hardware.
• Video of someone using and purchasing a Gama: https://www.youtube.com/watch?v=_uTkHJGkyDM.

Notes

• BBC Micro Model B (the one and only family computer)
• Sinclair Spectrum +2
• Atari 520STFM
• Amstrad PCW9512
• Atari Lynx II
• Nintendo GameBoy
• No-name DOS 386 PC
• Atari Falcon030 (yes, really)
• Apple PowerMac 8200/100
• Sony PlayStation
• Sega Dreamcast
• Apple iBook G3

…and the rest have been functional but boring MacBooks.