Sinclair’s 1984 big shot at business: The QL is 30 years old

Archaeologic In May 1983, Sinclair Research Managing Director Nigel Searle began briefing the press about the successful British micro maker’s next big release. It was 13 months after the company had launched the Spectrum and although that machine had become a huge success, punters and market-watchers were keen to hear about what Sinclair was planning for an encore.

Searle revealed the follow-up to the Spectrum would début early the following year, a date emphasised by the codename he gave: ZX83. The new machine might share the same naming scheme as the ZX80 and the ZX81, but it would be no enthusiast’s micro: set to retail for around £1,000, it would be a business machine. However, it was not an IBM PC clone - Searle said it would run a new operating system of Sinclair’s devising.

A dangerous ploy, some would have thought, avoiding both the slowly fading de facto standard business micro operating system, Digital Research’s CP/M, and its upstart rival, Microsoft’s MS-DOS. Lest potential buyers be put off by Sinclair’s desire to force buyers along its own, third way, Searle teased them with the notion that the ZX83 might even be portable, sport an integrated display based on Sinclair’s flat-screen TV technology and feature solid-state storage.

The machine Searle was busy dropping hints about would eventually be released as the QL - short for "Quantum Leap" - and it would indeed be launched early in 1984, on 12 January at the Inter-Continental Hotel on London’s Hyde Park Corner. Many more months would pass before the new Sinclair micro arrived in buyers’ hands, however, prompting wags to dub it the "Quite Late".

But it would not be a mobile machine. Nor would it have a built-in screen, or on-board solid-state storage. It would jump out of the 8-bit world of the Spectrum and its rivals, and into the 16-bit domain. It would also wave farewell to chip designer Zilog, creator of the ZX machines’ Z80 CPU, in favour of Motorola and its 68000 family. Out went Sinclair’s trademark single-touch Basic keyword entry too.

In charge of creating the QL was David Karlin, a young engineer in his early twenties who had just returned to the UK from a stint in Singapore with the Fairchild Camera and Instrument Corporation, where he’d been engineering digital signal processor chips. Back in Blighty – not entirely a move of his own choosing as it was his wife of the time who wanted to return - Karlin approached a recruitment consultant about a possible job opportunity he’d seen advertised. The advisor said the post wasn’t right for him and suggested instead a vacancy that had just come up at Sinclair Research. A meeting with Clive Sinclair later and Karlin was hired as Chief Design Engineer, Computers.

David was one of a number of bright sparks Sinclair Research brought on board on the back of rocketing Spectrum sales. Karlin wasn’t particularly keen to return to Cambridge, where he’d taken his degree and MA – in Engineering and Electrical Sciences – but Sinclair was willing to pay him the salary he’d been making at Fairchild, well in excess of what he might otherwise earn in the UK, so he couldn’t really say no.

Star hire, Star plan

Besides, he had a computer he desperately wanted to build. His time at Fairchild had begun with a brief induction period at the company’s Palo Alto HQ. There he’d encountered a Xerox Star, the colossally expensive workstation that introduced the WIMP (Windows, Icons, Menus and Pointer) user interface and which - unknown to him at the time - inspired Steve Jobs and Apple to build the Lisa then the Mac, and Microsoft’s Bill Gates to start work on Windows. David was enthralled, and like Jobs and Gates realised that here lay, in concept, the future of personal computing.

His notion, then, was to “create a £500 Xerox Star”, and it was this vision that helped persuade Clive Sinclair to give him a job. It also helped that Sinclair Research desperately needed to devise a business micro for recently denationalised UK IT giant ICL.

In May 1981, Robb Wilmot joined ICL as its new MD. He was just 36 years old, and in his previous role he had been Texas Instruments’ youngest vice-president, in charge of the UK calculator division. He was also pals with Clive Sinclair. Wilmot believed that Sinclair technology could enliven ICL’s desktop computing range; ICL’s Dataskill subsidiary was already offering programs for the ZX81 and had been making business-centric software for home micros since the Nascom 1 days. Sinclair and Wilmott cut a deal, announced to the public some month later, in December 1981.

First, ICL would license Sinclair’s version of Basic, a move likely suggested by Sinclair in response to the BBC’s decision to adopt Acorn technology for the BBC Micro. It had always been Sinclair’s line that the BBC should have partnered with him because he, not Acorn, had the closest thing there was to a standard dialect of the language – ICL’s support for Sinclair Basic would cock a snook at those Corporation fools.

Second, ICL said it would also buy Sinclair flat-screen TVs for a combined desktop computer-cum-communications terminal it had a mind to offer. Sinclair would develop the machine’s underlying hardware. In return, Sinclair gained £1m up front to help finance development costs. It also received a commitment from ICL to pay it royalties when the product shipped.

Sinclair Research’s industrial designer, Rick Dickinson, still has a series of concept drawings dated to between May and August 1981 which show various wedge-shaped machines equipped with a pair of Microdrives, the tape-based storage system in development at Sinclair, and what appears to be a small, integrated display. Some sketches also feature a built-in printer. The angular look presented in Rick’s pictures is not a million miles away - phone handset notwithstanding - from how ICL’s One-Per-Desk (OPD) eventually appeared.

There were many other ideas for future machines being kicked around at Sinclair too. Today, David Karlin recalls talk of an advanced, second-generation Spectrum that could also be used by business and educational establishments. He believes that’s what Clive Sinclair had in mind when he outlined his vision of a £500 Xerox Star. The Sinclair chief always had an eye on the business market, even at one point trying to suggest the ZX81 was a business micro.

Despite their apparent similarity to the OPD, Rick Dickinson’s sketches may also reflect general thinking within Sinclair Research about its post-Spectrum products. These concepts may even have persuaded Robb Wilmot to ally ICL with the company in the first place, and influenced the parallel evolution of the OPD look. Thirty years on, memories are too hazy to be certain. But then Sinclair Research never went in for formal development processes, or for consensus views on what a future machine would do and for what audience.

Planning the ZX83

Rick’s early design work was done more than a year before Karlin joined Sinclair, and while work was proceeding on the Spectrum itself. The ZX81 had only just been launched at that point. Karlin arrived in the late summer of 1982 and immediately began putting together a preliminary specification for his business machine.

He envisaged a desktop computer that would feature a proper keyboard and “some kind of networking”. It would come supplied with a dedicated monitor and its own printer, “both of which I considered to be really the minimum for a decent business machine”, he recalls. “In fact, in terms of broad spec, you could say I was trying to design what the Amstrad PCW 8256 would eventually become.”

He also wanted some kind of windowed user interface, which would in turn require high-resolution bitmapped graphics, plenty of memory and a fast processor. Intriguingly, though, he did not plan to add a mouse. “That seems an odd choice now, but at the time, I was fairly confident that people could do enough with arrow keys,” he says.

The early design work centred on the silicon. “I asked myself, how can I get high-resolution video, which was clearly needed, in a way that isn’t going to cost money. Costing money essentially meant costing pin counts. That was the critical issue at the time. When you built a custom IC, if you went into anything above a 40-pin dual in-line package, the cost skyrocketed. So as far as I was concerned I was limited to 40 pins, which mean a hell of a lot of gate-count arithmetic and pin-count arithmetic.

“The design style in Clive’s shop was, if it costs ten pence and we can get rid of it, then let’s get rid of it; if it costs a pound and we can get it down to 90 pence, then we’ll do that, and if we can get it down to 50 pence that’s better still. Literally, you were counting the cost of capacitors and the design process was all about getting every last inch of cost.

“In a sense my game was: I know the base minimum functionality I want, I know the cost constraints that Clive wants, so can I make my stuff fit into that constraint? And that was pretty much the design process.”

Since the ageing Z80A microprocessor could only address 64KB, and Karlin was adamant that paging blocks of memory in and out would hinder performance, he also decided that his machine needed a 16-bit processor. He considered Zilog’s Z8000 and Intel’s 8086, but quickly rejected both in favour of what was, for him, a much better option: Motorola’s 68000 series.

“It was clear the Z80 was going to have two big problems,” he says. “Firstly, it just not being fast enough and, secondly, not having the addressable memory space. The idea of faffing about doing our own hardware paging, which was the only option when you only had a 64KB address space, did not appeal.”

New era, new processor

Neither did the 68000 itself, but for a different reason. Instead, Karlin chose the 68008, running at 7.5MHz - twice the frequency of the Spectrum’s Z80A. The 68008 was built for backwards compatibility, so it featured an external 8-bit data bus and 20-bit addressing. The 68000, as used in the first Mac, the Amiga and the Atari ST had a 16-bit data bus and 24-bit addressing. Both Motorola chips could handle 32-bit word lengths internally.

“I don’t know how robust the decision was,” David admits now, “but it seemed fairly clear that the the 68000 series would be a great platform for the future - it was a very, very good processor. The problem with the 68000 was basically a pin-count issue. Motorola was pricing it gigantically higher than the 68008, double or treble the price. It was sufficiently high I didn’t even argue about it.”

Rivals’ use of the full 68000 would later come as something of a surprise. If Sinclair couldn’t afford the 68000, how could they? Today, David blames Sinclair’s negotiating skills, not just for the CPU but for a whole variety of logic chips and add-ons: “I question how good we were at purchasing, because people like Amstrad, certainly the Japanese, certainly Apple, who did not have gigantically higher volumes than us at the time, got massively lower prices.”

Based on Karlin’s recommendation, Clive Sinclair approved of the use of the 68008 in December 1982, formally pegging the company’s future on the platform. During 1983, it has been claimed, Motorola cut the price of the 68000 to below what Sinclair had agreed to pay for the 68008. Renegotiating the purchase contract might not have been costly, but adding in the architecture the full 68000 chip required - extra banks of Rom and Ram, a separate system logic chip, would have been, so it was decided to stick with the 68008.

Ironically, given the way the ZX83 hardware evolved through its development, it’s easy to say Sinclair would have well been better off going with the 68000 after all, but only with the benefit of hindsight.

Karlin decided that just two custom chips would operate alongside the CPU. The first, the ZX8301, would connect the 68008 to the rest of the system. It would operate as the processor’s clock, manage the memory timing, and mediate CPU and display controller access to Ram. It would also control the display through a dedicated link.

The second ULA, the ZX8302, would handle the IO, including the Microdrives, the network and a printer port. It would integrate a modem - mandated by ICL’s requirements for the OPD - plus the ZX83’s planned battery backed real-time clock and the keyboard interface. Apart from Ram - then imagined to be 64KB split 50:50 between program space and video memory - and Rom chips, that’s all the ZX83 would need, it seemed.

Naturally, it never quite worked out that way.

Work begins... but on what?

At the start of 1983, with the basic system logic outlined and a CPU selected, Karlin started to devise the machine that would be based on these components and to specify the core software that would be required. In March 1983, he was given a nine-month deadline to deliver the ZX83. This would allow the machine to be launched just before Christmas.

Even then it seemed, to some, a crazy deadline: Sinclair hadn’t yet succeeded in creating a machine in such a short timeframe, and this was no extension of a well-understood, established platform. Sinclair’s chief hardware engineer, Jim Westwood, appears to have been a lone senior management voice arguing that more development time was necessary, but his advice was effectively ignored.

The short development period would be tough enough for a new desktop machine, but it seems doubly arduous if Sinclair was indeed intending that the ZX83 be a portable computer, as Nigel Searle’s May 1983 comments to the press suggest. David Karlin says that the notion of creating a portable machine didn’t come until much later, but his chief software goal toward which ZX83 development initially proceeded during the first few months of 1983.

Sinclair bosses clearly did want to offer a portable, particularly following the May 1981 introduction of the Osborne 1. It’s very telling that the sketches made by designer Rick Dickinson around that time feature a machine with a small display, two removable storage drives and an integrated keyboard - the very defining features of the Osborne 1, though it’s impossible to say from the drawings whether they represent a portable or a desktop.

In April or May 1982, Clive Sinclair when questioned about what was then the ZX83, said: “The next step will be to make a machine of a suitably higher price [than the just-launched Spectrum] which would have a built-in screen and dual floppies - Microdrives, that is. It is conventional in the sense that it contains what the Osborne or the IBM personal computer have, because that is what is needed.

“We have three elements that people will want: our printer, the flat-screen display, which is critical - the world needs flat screens, that technology is paramount - and the microfloppy, and you bring them all together. That package becomes a much handier package than, say, an IBM system.

“We are doing something that is maybe a couple of pounds in weight - say two to four to be on the safe side.”

Was the ZX83 intended to be a portable, despite David Karlin’s desire to build a low-cost business desktop? Certainly there are clear parallels between his notion of a highly integrated system, and a portable design with integrated screen, printer and modem. A portable computer was clearly a topic of discussion at Sinclair long before Karlin came aboard, and would continue to be a feature of the company’s roadmap, such as it was, ultimately finding physical form in 1985’s ill-fated Pandora project. This bizarre machine, complete with a Sinclair micro-display, was killed when Amstrad took over, but it eventually led to Clive Sinclair’s Z88 portable.

Most likely, the portable was always Sinclair’s ‘next-but-one’ micro. Some seem to have certainly experimented with the idea: Tony Tebby remembers an Sinclair micro-display being hooked up to a Spectrum and the barely legible results prompting him to add proportionally spaced fonts to an early draft of the ZX83 OS as this might allow the display to present readable text.

One suggestion that the portable be powered by batteries produced for the Sinclair Microvision 2700 pocket TV was very quickly dismissed when calculations showed they would support just 30 minutes of operation, falling to a mere 10 minutes if the Microdrives were running.

If there was any serious plan in place to build a portable, it was wrecked by these findings. But such was the nebulous nature of the ZX83 project, with all participants seeming to have subtly and not-to-subtly different ideas about the machine they were building of having built for them, that at this remove it’s very difficult to tease out the "true" specification. One engineer’s research effort was another’s provisional spec.

Suffice it to say, certainly by the summer of 1983 but more likely rather sooner, anyone hoping that the ZX83 might be a portable computer was now sure it would not be. The lead times required for designing, sculpting and engineering the case were such that, to ensure a late 1983 launch, the machine’s form-factor would have to have been set in stone by the middle of the year.

Planning the software

While David Karlin was sorting out the hardware during the early part of 1983, Tony Tebby and new recruit Jan Jones were working on the computer’s core software: the operating system and its Basic interpreter.

An Oxford physics graduate specialising in nuclear and solid-state physics, Tebby originally set out working for GEC on microwave systems. His work exposed him to the protean computer-aided design software of the time, which he found crude and unsuitable. So he learned programming “at night school” in order to try and produce something better himself.

This led him to a career shift: into CAD coding, which took him first to Philips and, later, to a joint venture between ICL Dataskill and the Department of Trade and Industry: CAD Centre in Cambridge, which he joined in 1979. It was there he met Jan Jones, who had become a programmer straight out of Sixth Form. She studied for a Maths degree part-time while working at British Gas and, later, the Royal School of Mines. Then she joined CAD.

In 1982, Tebby was hired by Sinclair Research as software engineer, part of the company’s same grab for talent that brought David Karlin into the fold. It was Tebby who invited Jones to come and join him at Sinclair, to work on SuperBasic, a version of Sinclair Basic to be brought up to date with the addition of some of the structured programming features that had been built into BBC Basic. Jan remembers coding it all up in 68000 assembly language after a long specification process in which the language’s commands and features were mapped out.

David Karlin insists he always intended to build the language into the machine, not only for application programming and to allow third-party applications to be run, but also as the ZX83’s “elegant, easy to use, nice and fully featured” shell language.

Tony Tebby, on the other hand, recalls bringing Jan Jones on board when Sinclair bosses decided that the core ZX83 hardware might also form the basis for a future incarnation of the Spectrum.

Karlin also says it was always his intention to use Tebby’s QDOS operating system, called "Domesdos" during development, and that the software engineer’s work on it was not a "spare time" project as has been suggested. For that reason, it was once claimed - and the claim has continued to be repeated by others ever since - that QDOS was a last-minute substitution, dropped in when it became clear an OS being developed by an outside contractor, GST Computer Systems, would not be ready on time.

No so, says Karlin: “That’s the wrong way round.” GST was the insurance policy taken out by Sinclair management, he says. The thinking went: “If this his little in-house team we don’t know and don’t necessarily entirely trust don’t deliver, we’d better have an operating system up our sleeve, so GST got commissioned to do that.”

Playing it safe

Tony Tebby’s recollection is different. He says he commissioned GST, initially with no formal approval from Sinclair management, to create an operating system for the new machine. “I was brought in to find, not necessarily to write, the QL’s operating system,” he remembers. He started to author one anyway, though, just in case.

But he once wrote, “I would never have been able to take the technological risks that I did if I had not been confident that, however badly I screwed up, GST would produce the world’s best multitasking workstation OS on time”, which seems a tacit admission that GST was indeed the back-up.

Tony says he also tried to interest Digital Research and Microsoft in the job: DR simply wasn’t interested, and Microsoft said Tebby’s conception of what the new OS would be capable could not be realised on the hardware of the time.

Today, he chuckles at the famous firm’s seeming inability to do what both GST and he were, separately, able to achieve. David Karlin says Tebby’s work, with its incorporation of, for the time, advanced features such as multi-tasking, was far ahead of anything Microsoft would have been able to come up with.

How was Domesdos selected in place of GST’s work? Tony Tebby says his operating system and GST’s were put “head to head” at a meeting held in November 1983, though he admits he was not present. The result of this review, however, was that GST’s OS was out and his was in. Why? Most likely because, being an in-house job, Domesdos could more quickly be updated once the still-unfinished hardware became available. GST later released their work under the name 68K/OS.

“Conceivably, different people had different impressions of what was on,” says Karlin. “It’s completely possible that so far as Nigel Searle was concerned, Tony’s work was being done to keep me happy and the intention was to use GST and when that didn’t happen, we carried on with QDOS. That was not my impression, but that’s not to say you didn’t have different camps of people thinking different things.

“As far as I was concerned, the stuff Tony was doing was what was going to be used and that the GST stuff was there as a back-up in case we were unable to make QDOS work, but we were able to make it work.”

Separately, Psion software, which would go on to create the Symbian mobile operating system, was hired early in 1983 to produce a set of productivity applications which could be bundled with the new machine. Psion had been approached by Nigel Searle during December 1982, as had a number of other software houses with a view to sounding them out as potential supporters of the new micro. David Karlin says that including applications was always part of the plan, though he was not involved in choosing Psion. Neither was Tony Tebby, it seems: GST has a suite of its own in the works alongside its work on the OS.

Applications, applications, applications

Quite possibly Psion’s very ambitious Managing Director, David Potter, cut a deal early on to ensure Sinclair funded the development of Psion business applications which would be initially released on the ZX83 but could later be released on other platforms. Alun Sugar, in his autobiography, calls Potter and co. “an arrogant bunch of tossers”. Tebby claims the company as conveniently already working on MS-DOS applications which it could re-purpose for the QL deal. He cites a late-in-the-day request that the QL should support the 80 x 25 textual display for which the apps had been written, though this could easily have arisen because, with no ZX83 hardware, Psion had had to guess the machine would use this then common screen format.

Potter, like many others in the micro business at that time, realised that the market for business computing was about to grow the way the home computing market had been, and wanted to get a foot in the door. The ZX83 project would certainly have allowed Psion to do so with much less risk than launching standalone applications would. If the ZX83 sold as well to business as the Spectrum had to teenagers, it would establish Psion’s Quill word processor, Easel graphics tool, Archive database and Abacus spreadsheet – together later called the xChange suite - as a new de facto standard in 16-bit business applications.

“We had long discussions with Sinclair,” Potter said in 1984. “Psion is an ambitious company... We wanted... to be producing the fundamental software tools - the word processors and spreadsheets - a market dominated by the big US software companies.”

Psion devoted almost all of its coding resources to the ZX83 applications. “It has been a huge effort for us,” said Potter. “Frankly, this is why we haven’t been coming out with too many home computer products recently.”

Each of the four applications were put in the hands of their own team leader. Martin Brown was in charge of Easel development; Martin Stamp oversaw the Quill team; Charles Davies ran Archive development; and Abacus’ was Colly Myers’ baby. Myers went on to become Psion’s MD and, later, CEO of Symbian.

Psion’s programmers would spend the next 15 months completing Quill, Easel, Abacus and Archive. With only a very basic idea about how the ZX83 would operate, they coded the applications on the company’s Vax minicomputer running a 68008 emulator until the ZX83 and QDOS were complete.

But toward the end of 1983 neither were. The very fluid concept of what the kind of micro the ZX83, and Sinclair management’s refusal to extend the development deadline, were to blame. Deciding that the business-centric ZX83 platform might form the basis of future Spectrum home computers didn’t just mandate a new, 68008-based version of Sinclair Basic, it also meant that David Karlin’s initial specification for the machine’s two ULA chips had to be revised.

For instance, the ZX8301’s display controller circuitry would need to be altered to provide support for existing Spectrum display modes. The ZX8302 would now require a slick sound generator, to allow the machine to produce rather better sounds and music than the Spectrum could.

Separately, it was decided it would take too long to build a modem into the ULA so it was dropped, along with with output-only printer port, to be replaced by two generic RS232 ports.

Shifting specifications, fixed deadline

Getting all the extra functionality in under the constraint imposed by the chips’ ‘40 pins maximum’ limit proved almost impossible - the serial IO now required eight pins not four, for instance - and this appears to be the main reason why plans for a ZX83-based Spectrum were quietly shelved late August or early September 1983. With the ZX8302 still unfinished and “running out of pins”, as Tony Tebby puts it, it became necessary to offload the keyboard interface.

Enter one the ZX83’s more controversial design decisions: the use of an Intel 8049 microprocessor in the system logic. It was dubbed the “Intelligent Peripherals Controller”, something of a misnomer. David Karlin put it to work handling the keyboard input. “Again it was all pin-count driven,” he explains. “To be able to do the keyboard we needed something with lots of pins to drive all these keyboard lines. So we asked, what’s the cheapest thing we can find that has lots of pins? It turns out that it’s not a few bits of TTL you drive directly off the main processor, it was actually cheaper to have a second processor.”

Especially since the 8049 would have capacity left over to deliver the sound generation and to free the ZX8302 from having to handle the serial ports. There were downsides: Karlin originally envisaged the ZX8302 providing a centralised bank of interrupt and status registers for peripherals, a smart scheme for reducing hardware costs and simplifying the software. Adding the 8049 meant this was no longer possible - the 8049 had to use its own status registers.

Tony Tebby handed the task of interfacing the 8049 into QDOS to a new assistant, Aaron Turner. It was a tough job, but Turner managed it. It was much, much harder getting the 8049 and its lines onto the long, narrow PCB mandated by case, which had been agreed at a point when the specification was such that everything would fit.

The PCB size problem was compounded by a relatively late move to make the ZX83 itself more friendly to home users: it would now have to have TV output and joystick ports, odd additions for a business micro. The joysticks could be added with a pair of ports and a few lines running off the 8049, but adding a UHF modulator was a major headache. It could only be fit in right next to the Microdrives – with their frail read/write heads, this was the last place you’d want to put a source of oscillator noise.

And if the ZX83 was to appeal to home users, it would also need a Basic. Jan Jones was hard at work on SuperBasic, which would fit the bill. The trouble was, SuperBasic was not only scheduled separately - designed for a post-ZX83 Spectrum-brand product, it was months away from readiness - but it had not been designed to run on top of QDOS. Instead, it was a classic ‘boot into Basic’ operating system in its own right, says Tebby, as per the Spectrum and almost every other 8-bit micro.

Adapting SuperBasic’s graphics routines – then under development at GST in a bid to bring the release forward – to operate through QDOS’ display manager rather than write direct to the video Ram was relatively straightforward. Aaron Turner was given the job. Rewriting the interpreter to obtain or release memory through QDOS’ memory manager rather than do so itself was a much more involved task.

And one that would have taken too long if Tebby and Jones were to meet the pre-Christmas launch schedule. With just four weeks to go, much of which had to be reserved for testing the code with hardware, Tebby elected to hack QDOS and SuperBasic to allow the interpreter to run as a special, one-of-a-kind privileged task. It was, he says, “a terrible mistake [that] compromised the OS’ integrity”.

A tale of two Basics

At this stage, it seems to have been still the plan to deliver two versions of SuperBasic: a compact, minimal version in the ZX83’s Rom, and an extended version which comprised the full language but would come later, on Microdrive tape or on a plug-in Rom cartridge. The minimal SuperBasic would check for the presence of Extended at start-up and load the extensions accordingly. This would, says Tebby, allow the ZX83 to ship with Basic and still give the coders time to finish the full package.

Not that any of the hardware concept changes - from portable to desktop, from ‘Super Spectrum’ platform to an ‘Spectrum plus’ in its own right - had warranted any extension of the development team’s deadline, at least not as far as Sinclair management were concerned.

The only formal set-back to the timetable was to schedule the launch the machine early the following year rather than before Christmas. David Karlin and Jan Jones both say the machine’s development was then at a stage where it needed at least six more months’ work, a duration defined, in part, by the then long turnaround times involved in producing PCB and ULA samples. According to Tebby, they didn’t even have a working prototype at that point, almost entirely thanks to the many changes that needed to be made.

Karlin would surely have made Nigel Searle and Clive Sinclair aware of this, but nonetheless the decision to introduce the machine to the press in January 1984 was upheld. This annoyed Tebby, but the claim made at the launch that the first machines would be in buyers’ hands within 28 days of their orders being received was the final straw: knowing this deadline could not possibly be met, he handed in his notice and said he would leave Sinclair when there was a machine in production.

He duly left Sinclair’s employ at the end of April, though he was set up in an temporary office round the corner, he says, where he could be called upon should his assistance be required. He refuses to name the Sinclair Research director who arranged this.

The launch had another effect: the inevitable Sinclair miscommunication ensured that the company’s marketing team knew only that the QL, as the computer was now called, would come with a Spectrum-friendly Basic, not what it could and couldn’t do, or that there would be a two-stage release. They assumed they could fill out the pre-release documentation for the press with information from the Spectrum manual. But they didn’t check it with the developers. The upshot was that, having told the world the QL would ship with what was effectively the full version of SuperBasic, the company would now have to supply it.

At the launch, Searle promised Sinclair would begin taking orders for the QL at the end of January. The computer would ship with 128KB of Ram, and be priced at £399 plus £7.95 postage and packing. There was no mention of the 64KB version, originally intended to go on sale for £299, because Psion’s applications would need more than 32KB of memory. The machine would make is retail debut the following Autumn, Searle said, when it would also go on sale in the US for $499.

Of course, Sinclair hadn’t managed to get its previous computer, the Spectrum out on time, and the Microdrives took even longer to arrive: they were more than a year late. A few observers rightly guessed it wouldn’t do so this time round either. But such fears didn’t stop many thousands of punters - more than 9000 by the end of February, rising to 13,000 as of late April - sending off their credit card details or their cheques, which the company had no hesitation in cashing.

Delay after delay

That the QL subsequently failed to appear as promised was bad enough, but worse, customers were out of pocket. In an attempt to cap the growing flow of bad publicity, not to mention the growing interest in the saga of the Advertising Standards Authority, Sinclair promised to put the money into a “trust fund” and pledged not to touch the cash pile until the first QLs were despatched.

That didn’t come as much relief to waiting buyers, who quickly calculated how much interest the account would generate for Sinclair while they were left waiting. Sinclair at least allowed them to cancel their orders and get their money back. But very few seem to have taken the company up on its offer.

By the middle of March, eight weeks after Sinclair began taking orders, no QL had shipped. Officially, QDOS was taking longer to finish than expected and one of the QL’s two ULAs required further modifications. The latter claim is probably true, but Tony Tebby insists QDOS was ready to go and able to be committed to Rom along with SuperBasic in March 1984.

And they would not, he adds, have required the infamous QL dongle.

According to a Sinclair spokesman, the QL’s firmware was too big for the Rom. The QL could take 32KB of Rom internally, and QDOS and SuperBasic now took up 40KB, mandating a 48KB Rom chip.

Rubbish, says Tony Tebby. The QL was always specced for 64KB of Rom and, from January 1984, the PCB design incorporated two Rom slots, each able to take 8, 16 or 32KB Rom chips. The machine ought to be able to have a 16KB and a 32KB Rom chip fitted. Why, then, bother with the dongle, an inherent admission that the machine was flawed?

Because Sinclair felt it better to send out QLs that worked, sort of, than to delay the product even further in order to await fully working ones. It was easier to blame failures on software than it was to confess that the hardware still wasn’t ready. Duff software tacitly implied a better version in the works. Not wanting to accept that they’d coughed up for duff hardware, punters naturally believed the white lie and didn’t bat an eyelid when Sinclair did in fact replace their machines in due course.

‘White lie’? It seems that way, because the first QLs shipped with an early, pre-release, test version of the OS, ‘FB’, even though a working version, ‘JM’ had already been completed by this point. The dongle conveniently indicated which returned machines should be replaced rather than repaired.

Sinclair Radionics and now Sinclair Research had been adhering to a theory that it was better to ship and fix problems by sending out replacement products, than by delaying a product to get it right. Sinclair companies had never gone in for much in the way of pre-shipment production development; once a devices was out of the labs it was considered ready to build and ship. Only then would production improvements be sought.

Driving toward completion

March 1984 came and went and with it Sinclair’s second, revised deadline for getting the QL into buyers’ hands. By now punters were being told they would receive their machines by the end of April. Sinclair made it - just. It was claimed only a few dozen QLs shipped on Monday, 30 April, every one delivered to their new owners by taxi or Sinclair staff driving hire cars. Buyers who had had to wait were given a £15 RS232 printer cable for free.

A Sinclair spokesman called this a “goodwill gesture”, but irate buyers felt it was the least the company could have done given not only the delays but how buggy the new machine now appeared to be - thanks to early draft hardware and the aforementioned inclusion of a test version of the OS. “Shoddy finish and unloadable software seems to be the least of their problems,” wrote Your Computer in its June 1984 issue (published in May 1984). “The Screen Editor can make the system crash and the promised real-time clock is missing - along with the manuals.”

The RTC would never appear: the ZX8302 chip, one of the QL’s two ULAs, would have had to be modified to cope with a 68008 bus access bug which manifested itself during resets. It was easier to drop the battery and stop mentioning that the QL had a real-time clock.

David Karlin is tactfully restrained when it comes to discussing his personal feelings about the QL development process - except in his criticism of the Microdrive, which is the part of the system that caused the most headaches and, for him, ultimately sank the machine. He has a point. Most of the complaints raised by reviewers and users centred on the unreliability of Sinclair’s tape-loop system. The Spectrum version was finally released a year late, in August 1983. The version in the QL would need much improvement if it was to be seen as a viable alternative to floppy disks.

According to Rick Dickinson, “the Sinclair way” of doing things meant that rather than redesign the Microdrive internal chassis for the new computer, the Spectrum drives were simply put into the QL case.

“With different materials it would have been more reliable,” he says. “I wanted to start off with a brand new chassis in a different material and build the two microdrives on that, but they insisted that now we’d already tooled up a chassis, we’d just have to use that. And of course what would happen is that, because you’re going into a whole new case with a whole new set of parameters, you’d screw the microdrive chassis in and the torque reaction from the screws would twist all the boxes slightly and just slew the whole chassis - both of them - off at one degree.”

And with the mechanical design set in stone, all Sinclair engineers could do to enhance the Microdrives for the QL was to devise a new way to arrange data on the tape, up the speed and adjust the electronics.

“We needed to design a Phase Lock Loop (PLL) circuit to decode the magnetic signal that’s coming off microdrive and track speed variations and such,” recalls David Karlin. “Mixed analogue stuff was difficult and expensive and didn’t mix with some of the other gate-count requirements of CMOS technology, so I opted to go for a digital PLL. I had a conversation with Ben Cheese, the analogue engineer doing the Microdrive, and I asked what’s the duty cycle of the waveform how far away from 50 per cent will it be, and Ben said it’ll be very close to 50. So I went and designed a digital PLL which would work fine if the duty cycle was between 45 and 65 per cent.

“But that wasn’t right. The duty cycle could actually vary between 10 and 15 per cent rather than just five per cent. So apart from the mechanical problems inherent in the Microdrive anyway, the duty cycle was not matched to the PLL and the only thing that could be done by the time this was to try and really sort out the analogue electronics to get the duty cycle to 50 per cent. Ben did a heroic job, but it was a losing battle, so we ended up with an unreliable mass storage system and in my view that’s what killed the product.”

Reaching ‘QL 1.0’

The other issues were largely solved during the first six months of 1984, either with hardware updates – production specs were changing on a monthly basis, though the PCB was updated only a few times, up to Issue 5 – or through software workarounds. Not every fix was successful: beating a glitch on the ZX8301 which emerged when it got hot minimised the bug’s impact on Microdrive usage but at the cost of killing network compatibility with the Spectrum’s Interface 1.

By July, Sinclair was no longer having to bundle a Rom dongle – the QL’s now coming off the Datatech production line were build ‘D06’. Come the end of the month, Sinclair said it would soon begin asking for original, dongle-supplied QLs to be returned, free of charge for a Rom swap. The old machines were junked and simply replaced with D06 or later QLs.

“Our intention is to stagger the recall of machines, and as yet we do not know how long customers will be without their QLs when recalled,” a spokesman admitted. The ‘Rom-swap’ recall process began in August, by which time the company reckoned the turnaround time would be ten days.

By this point, many different versions of the language had shipped, internally dubbed ‘FB’, ‘PM’ and ‘AH’. ‘FB’ was the version that went out on the first QLs. ‘AH’ was, a spokesman said in July 1984, the “final version”. It wasn’t, says Tony Tebby, though it was the point at which there would be no new additions made to SuperBasic, only bug fixes. The true ‘final’ version of the firmware was ‘JM’, which was tested and ready by March 1984, Tebby says, though it wasn’t used in shipping machines until July 1984.

‘JM’ was followed after a few weeks by ‘TB’ and soon a new development version, ‘JS’, was established, after which Tony Tebby finally left Sinclair. It wasn’t ready for release, yet it formed the basis for US QLs because, claims Tebby, Sinclair staffers couldn’t find the ‘TB’ source code. ‘JS’ was ultimately followed by ‘MG’, which formed the basis for the first non-English QLs, in 1985.

In the Autumn of 1984, Sinclair began releasing the QL to its retail partners. By now the production spec was at build ’D14‘, featuring the latest motherboard, Issue 6, which added an extra TTL chip to cope with issues found in the ZX8301 ULA. The QL hardware was at last in a ‘1.0’ state. Future build updates were made to improve yields not tweak the basic design. The manual, largely written by Roy Atherton of Bullmershe College Computer Centre, but incorporating material by Sinclair staff programmer Steve Berry and Psion’s Dick de Grandis-Harnson, was now complete too.

The TV output’s high-frequency oscillator had been moved away from the left-hand Microdrive’s head amplifier to stop interference causing read problems that could, at worst, render the drive inoperative. On the drive itself, a capacitor had been placed in parallel with the head to block noise induced by the drive’s motor. The 1980s phone-style connector used for the joystick and serial ports was replaced with a nine-pin D-Sub: male for the joystick ports, female for the serials.

Getting the QL into shops allowed Sinclair the luxury of essentially relaunching the QL. It began booking television ad space for a commercial designed to establish the QL as a much more economical alternative to its key rivals: the IBM PC, the Mac and the BBC Model B, the latter priced up from £399 to £1632 by adding a pair of 5.25-inch floppy drives, a monitor and more. By contrast, the QL, pitched with the colour monitor David Karlin always wanted it to have came to £698.

The Microdrive mistake

And yet, there were still issues with the machine - most, but not all, due to its storage system. “The manager of the local branch of Dixons told me that out of 1000 machines delivered to their warehouse, only 190 worked properly,” claimed a Sinclair User journalist in November 1984. “Further rumbles from Spectrum distributors seem to indicate similar troubles, with one hapless dealer spending a whole morning with six QLs and six sets of Psion software trying to find a combination that allowed all the Psion wares to be loaded.”

One of the QL’s two ULAs, the ZX8301, contained the display generator but no buffering on the lines to the RGB monitor output. Yanking the screen cable while the computer was powered up could damage the chip, which then had to be replaced. Not so much a bug, this, more a product of keeping costs low. And was it unreasonable to assume that most users would leave their displays connected? That said, the TV output’s timings were off, causing overscan issues.

“The one really bad bit of cost-cutting I did was trying to squeeze two serial ports out of one,” David Karlin concedes. “That was not a sensible thing to do, and we should have just said it had one serial port. It was only one serial port, so the two we claimed had to be multiplexed and that was seriously not a good idea.

“Because you’ve only got an 8-bit bus - pin count issues again - you have to slow the main processor in order to allow the graphics [circuitry] to get at the memory... But if you put your extra memory in, then the memory that wasn’t shared with the video bus was suitably fast. It was really only if you had the version with just the one bank of memory chips that the video slowed the computing down to a disappointing degree.

“It was undoubtedly the case that the 68008, by the time you’d doubled up one the bus cycles and you’d taken a whole load of cycles away to run the memory, it was slower than we’d have liked. But those are tiny issues - what really mattered was that we didn’t have a reliable mass-storage method.”

Or one that software developers were keen on. In May 1984, Nigel Searle talked to the press about a Sinclair charm offensive aimed at persuading software developers to support the QL. Psion was already on board, of course, and Searle said he was now talking to Quicksilva, Melbourne House, Ultimate and Picturesque in the UK, Digital Research, Microsoft, Lotus, Ashton Tate and Software Arts in the States. He said he wanted 50 titles out by the end of 1984. When that date came round, there were actually fewer than ten.

Even had the Microdrives been reliable, only Sinclair could produce them, and they were proving far less robust than cassettes when it came to high-speed duplication. Making them wasn’t easy - in a May 1984 interview Searle said Sinclair was then punching out 100,000 a month with an eye to ramping up to 40 million a month at some undefined point in the future. It almost certainly never achieved that.

Software scarcity didn’t help sales any, and nor did the reputation for fragility and instability gained by those early, prematurely released models. A WHSmith spokeswoman said at the end of 1984 that sales had been “very slow” thus far; “disappointing” was the word used by a Boots spokesman. Estimates in the press put the number of QLs in users’ hands at just 40,000, a fraction of the machine’s potential audience. None of Sinclair’s promised add-ons, among them a 512KB memory expansion module, a hard drive interface and a modem, had yet materialised.

Moving on

At the start of 1985, David Karlin decided to leave Sinclair Research. Tony Tebby had made good on his promise to leave when the QL was in what he regarded as a fit state to ship to customers. Jan Jones, by then expecting her first child, left soon after.

Having build a micro from the ground up, Karlin now wanted a fresh challenge and a more commercially oriented one. He wanted to go into business for himself, he says, but was persuaded to stay by Nigel Searle, who put him in charge of manufacturing. He found it in a poor state, he says. The company owed money to its key contract manufacturers, Timex and Thorn-EMI Datatech - the latter, with Samsung, was also producing QLs, and its cashflow was precarious. Karlin spent the next 12 months making sure the bills got paid and the right number of orders placed.

Karlin replaced Production Director Dave Chatten who, in March 1985, was made joint MD with Bill Jeffrey. Chatten was a long-time Sinclair man, but Jeffrey was an outsider, having joined the company from Mars Electronics at the start of the month. Nigel Searle was sent to the States to run Sinclair’s US wing, which he had himself established in 1980 before being brought back to the UK in 1982 to take over day-to-day operations here.

But Searle didn’t take the QL with him straight away – Sinclair formally announced it was again delaying the machine’s US debut, originally planned for the previous Autumn. The management reshuffle had been prompted by Sinclair Research’s ailing finances. Its income had been hit hard by the slowdown in home computer sales and the costs of Sir Clive’s Wafer Scale Integration and C5 projects. By the summer of 1985 it looked like media tycoon Robert Maxwell might buy Sinclair for £12 million. The deal fell through: thanks to a massive product order commitment from Dixons, Sinclair’s desperate need for cash was postponed.

Now there was even talk of a new QL and the idea of a new Spectrum based on QL technology was revived, though the machine behind the rumours in the press was the upcoming Spectrum 128, an upgraded Z80A-based machine with 128KB of Ram and souped-up sound, all developed with Spanish money.

ICL’s One Per Desk had already arrived too, early in 1985, following a November 1984 launch. ICL took the core QL, complete with a pair of Microdrives, built in a modem, a phone and an answering service, and pitched it at business for £1195 a throw. But that’s another story. For now, it’s enough to say that in January 1986, the OPD won a Recognition of Information Technology Achievement award for the Systems Innovation of the Year. The win echoed the QL’s victory in the British Microcomputer Awards the previous July as Microcomputer of the Year. Some folk, at least, could look past the QL’s early problems to see its potential.

But the potential was never realised, primarily thanks to the inclusion of the flawed Microdrives. Even the much-criticised keyboard doesn’t appear to have put people off the way many an early reviewer feared it might. The QL 2 never appeared and neither did a rumoured business machine codenamed ‘Enigma’ or possibly ‘Tyche’ which was claimed to run Psion’s xChange out of a Rom that also included the Gem GUI.

If this machines ever existed as anything more than a broad concept being aired at Sinclair Research, it was cancelled in April 1986 when the company’s finances finally hit rock bottom and Sincair Research was sold to Amstrad for £5 million, less than half what it was considered worth the previous summer. One of the first things new owner Alan Sugar did was knock the QL on the head. Quite apart from the QL’s sales performance to date, Amstrad was doing very nicely thank you selling its low-cost, business-centric, highly integrated PCW 8256. Sugar also made lots of Sinclair Research people redundant, among them David Karlin.

By this point, some 139,454 QLs had been manufactured, at least 122,793 by Thorn EMI Datatech for the UK market and 16,661 by Samsung for Europe and the US.

Life after death

But the machine had its fans. One was a Stevenage-based company called CST run by David and Vic Oliver, who had set up shop making QL add-ons. With engineer Graham Priestley they devised a machine called Thor that was based on QL motherboards they found they could obtain directly from Samsung. They also tweaked QDOS to support the machine’s one or two floppy drives – configurations priced at £599 and £699, respectively. There was a 2OMB SCSI hard drive model too; it cost £1399. The Thor was quickly dubbed ‘The Son of QL’ in the press.

Citing its ownership of the QL intellectual property, Amstrad soon forbade the use of the QL name and QL components. There was briefly talk of an attempt being made to buy or license those rights from Amstrad, but there’s no sign that Amstrad was persuaded to part with them. Psion was more forgiving, and licensed Quill, Easel, Abacus and Archive to CST, its Danish sales partner Dansoft and its UK distributor Eidersoft.

The Thor went rapidly through a number of configurations before the Thor 20 appeared with an entirely new motherboard: clearly, CST was successful in persuading Samsung to produce compatible kit. However CST pulled it off, it demo’d the Thor at various computer shows in 1986 and 1987, and is believed to have even shipped a small number to buyers, all without interference from Amstrad. But too few were sold to allow CST to continue as a going concern and production appears to have ceased in 1988.

Separately, Tony Tebby designed a second-generation QL with the help of ex-Sinclair engineer Jonathan Oakley. After leaving Sinclair himself, Tebby had formed QJump, a software company dedicated to producing QL apps and utilities. He found potential backers for the project, but when it came to the crunch, they couldn’t or wouldn’t stump up the £250,000 Tebby needed to realise his design as a shipping product.

Tebby maintained his connections with the QL world, initially offering add-ons for Q-DOS and SuperBasic, and later recreating Q-DOS for the 68000-based Atari ST as SMS2. This was later re-released as SMSQ for Miracle Systems’ QXL, a 68000 add-on board for PCs. He is still a figure in the QL enthusiast scene.

After leaving Sinclair shortly after Tebby did – she was expecting her first child – SuperBasic writer Jan Jones devoted herself to raising her family. She’d always had a love of writing - she penned a book about SuperBasic – and alongside looking after the kids, she took up short story writing. She went on to become an award-winning writer of romantic fiction.

David Karlin took his redundancy money and founded the business he’d wanted to start in 1985, hardware company Alfa Systems. It created the DiskFax, a rather popular (with spooks) gadget for quickly duplicating floppy disks over phone lines. He went on to run audio company Harman International’s UK operation and was later a Managing Director at accountancy software company Sage. These days he runs Bachtrack, a global search engine for live classical music events. ®

The author would like to thank David Karlin, Tony Tebby, Jan Jones, Rick Dickinson and Jonathan Oakley for their kind help in the preparation of this article. QL aficionado Urs König has a site devoted to the machine’s 30th anniversary, here