Constructing a personal computer
Some 16KB of Arnold’s 64KB of RAM was dedicated to the screen and 5KB for the OS, leaving 42KB for the user. The 32KB ROM was segmented into two 16KB blocks one for the modified Mallard Basic, the other for the new code, though the Basic overlapped a little, Clayton explains today. Paging hardware flipped the 16KB section in use at the time in and out of the Z80A’s memory map in alternation with the video memory. The Basic interpreter was modified so that the part held in the second 16KB of ROM wasn’t required when the first was page in.
The screen memory was used to hold the display data according to the three modes Arnold provided: numbered 0, 1 and 2. The latter was the hi-res 640 x 200, two-colour mode, with the resolution falling to 320 x 200 and 160 x 200 in the other two modes, though the number of usable colours, selected from a palette of 27, rose to four and then 16.
The three modes could incorporate text too, at 20 x 25 characters, 40 x 25 and 80 x 25, respectively. One neat touch: swapping one of the 27 colours in and out of the available palette of 16, four or two caused the on-screen colours to change immediately, allowing some clever colour tricks to be performed in software.
“The monitor is a pixel VDU - the simplest sort,” said Chris Hall in 1984. “When you do the sums you find the Amstrad has to be the way it is. It’s a straightforward 16KB of memory for the display... You can ‘suck’ 16KB from RAM just fast enough to refresh the screen.”
Without working hardware, the Locomotive team had to make certain assumptions about what the machine’s capabilities would be. “We had an idea of what price the machine was going to be sold for, which set the parameters for the hardware,” said Hall. “We couldn’t knock out something ‘noddy’. On the other hand, we didn’t have time to produce a QL from scratch."
Booting up: the CPC 464’s initial screen
Hall adds: “There were a number of constraints which came from Amstrad’s original thinking: the casing and keyboard and the inclusion of a cassette recorder and a monitor. That was all settled and had to be taken on. A Centronics port seemed appropriate for a printer. Having a built-in cassette player made life easier because its electronics are a bit special. Add a custom gate array and there you have the machine.”
That was the agreed specification, and that’s what Mark-Eric Jones and fellow hardware designer Roger Hurrey worked to create while the Locomotive team were writing the software. Jones worked on the main board, Hurrey on the gate array. Hurrey was another one-time Data Recall colleague of Jones, Clayton and Hall, and had in fact designed the ill-fated IBM PC accelerator board under contract to Locomotive.
Mej remembers the specification process: “Amstrad was basically saying, ‘We want to get this thing into production quickly, and we have certain key requirements’. They decided for marketing reasons, the most important thing was to say that it had 64KB of memory and that it had lots of colours. They threw in one hand grenade: they said, ‘Oh we’ve actually ordered quite a lot of the components, and if you can use the same components that would be very helpful’.
"In the end we used a very different set of components but helped solve the commercial problem of having already placed orders with suppliers by getting them to order from those same suppliers.”
The original 6502-based board didn’t incorporate any custom silicon. “We decided to use a gate array to make a lot more functionality available than we could have done otherwise. By having a lot of logic - about 1200 gates - embedded into a gate array, it had a lot more functions that it could possibly have had in the original version which was shown to us and which didn’t have anything like that. That allowed better video, way beyond what they were proposing for the original board.”
Managing the contractors
There was a lot of pro-active hassling, remembers Roland Perry. “I’d go out on a Friday evening and pace up and down on someone’s premises and say, ‘Have you finished? I want to this ROM or this chip layout.’ And they’d say, ‘We’re not quite finished yet, can we deliver it to you at nine o’clock on Monday morning?’ I’d say, ‘No, because the person who is going to work on it next is starting at nine o’clock tomorrow!’ Our timetable was so tight everyone was working weekends.”
Perry had done critical path analysis work at University, and this he now applied to help him map out a timetable for his two contractors, Locomotive and MEJ: “If we throw everything we’ve got at this, and if the consultants give it top priority, and we schedule it down to the wire, we can get it done.
“The rush was because of the annual cycle of Amstrad product development. Amstrad wanted to have product in the shops at Christmas. If they’re going to be in the shops at Christmas, they have to be on Dixons’ shelves in September, which meant they had to be on the boat from the Far East six weeks before that, which meant they had to be being built in the factory a month before that.
“We had to have a timeline that meant you could launch the product around Easter, which meant you had to have a reasonable number of working prototypes, preferably built in the proper factory, which meant that the people with the factory have got to have built a production line, and they wouldn’t do that unless they knew that it was going to work and not vapourware. So drop dead time was the first week in December, when you went to the Far East and showed them what you wanted them to build.”
“It was a time of intense work,” recalls Mark-Eric Jones. “We were living off frozen pizzas more or less, and every evening working late into the night. This was largely happening in the house I owned at the time. Roger was there with us. We had a drawing board out. There wasn’t any schematic capture back then so it was all drawn with pencil on paper, and that’s how the design was evolving. We were building prototypes in the spare bedroom. My wife – she was my girlfriend at the time – was hauled in to solder prototype PCBs.”
Jones and Hurrey made rapid progress on the Arnold hardware. Locomotive got to look at the the first working hardware at the end of October 1983, though at this stage the machine’s ULA chip was not ready so it had been replicated in circuit board form so the software guys knew exactly what it could and couldn’t do, and could test working hardware. This board, designed and built by Hurrey, was dubbed the "Gate Array Simulator" – or “the Gas Board”, for short. The PCB layout was made, from Mej’s schematics, either by Alan Jones, a contractor the engineer had used for many other circuit board projects in the past, or one of Ambit’s contractors.
The ULA would not be completed until February 1984, but the PCB layout was done and dusted during November 1983. The following month, with the Basic and firmware finalised but for four weeks’ bug spotting and squashing, Roland Perry and William Poel began sending out hand-built Arnold prototypes, each a laden PCB with the Gas Board riding piggyback and with a small doohickey comprising 32KB of battery-backed RAM to hold the firmware in place of expensive EEPROMs, to key software houses. Alan Sugar had insisted that the machine would need to have plenty of games ready at launch, perhaps as many as 50.
Getting working kit out to developers nice and early was a masterstroke. Where other manufacturers didn’t tell developers about new machines until the hardware was formally launched, Perry and Poel’s approach put working kit in coders’ hands and quickly won Amstrad the thumbs up from almost all of the major games developers. In turn, that meant a ready supply of software when the computer finally went on sale. It also ensured Alan Sugar would have games to show when the computer was launched.
Perry took on Sally Tyler to manage the logistics of getting these confidential prototypes out to developers and away from prying eyes. She was almost certainly successful, but nothing could really prevent the odd leak. In January 1984, Popular Computing Weekly reported that it had heard that Amstrad will offer a sub-£200 micro late the following Spring as “a direct competitor for the new Elan”. The report said the new machine would have 64KB RAM expandable to 4MB, include a full-travel keyboard with separate numeric pad, sport “similar graphics to the BBC” and have a three-channel, six-octave sound system. It would have a Centronics interface but no RS232, the paper reported. Its claims were surprisingly accurate.
For its part, Amstrad wasn’t exactly keeping mum either. Alliterative marketing manager Malcolm Miller told Your Computer in February that the company was working on a new machine and that “what we’re going to offer will be extremely successful [because] we have tremendous experience in marketing”.
Or, rather, the sales skills applied by Amstrad senior staff to persuade major high street retail chains Boots, Comet, Dixons and Rumbelows to commit to buying 100,000 machines in total. Combined, those orders underwrote Amstrad’s production and development costs.
The machine’s PCB layout, the software and the casing mouldings were all sent out to Amstrad’s production partner in South Korea, Orion, and with the ULA signed off early in February, chips could be sent over for assembly too. Orion would bill Amstrad cost plus five per cent for each machine that rolled off its assembly lines. With everything ready in March 1984, there was time for production to be tweaked for greater yields before the machines had to be in the shops.
Well, almost. According to Sugar, Ferranti, the leading British manufacturer of custom chips and the firm selected to produce Arnold’s ULA, let them down “at the eleventh hour”. Roland Perry says the problems centred on the semiconductor process Ferranti used; the ULA design was attempting to achieve two things for which the Ferranti process was not really suitable. There was nothing wrong with either process or design, they simply didn’t mesh on two of the chip’s functions.
Acorn ran into similar bother when it was designing a monumental ULA for the Electron.
“The layout of the gate array was done manually by draughtsmen putting tape onto plastic film to draw all the tracks for the chip itself, using a scalpel to cut the tape,” remembers Mark-Eric Jones. “In the first chip we had, the outputs were completely non-functional. It turned out it was because someone had forgotten to connect the IO pads around the side to the logic in the core.”
In those early days of gate arrays, there were no electronic design automation tools or digital design simulations to help chip designers, at least not in the UK. Debugging the ULA was an entirely manual process.
“I was catching the shuttle up to Manchester airport so many times, one of the taxi drivers there recognised me,” says Mej.
Fixing the issues Amstrad’s team had uncovered would have involved either redesigning part of the ULA – far too time consuming – or modifying the motherboard with the addition of a lot of extra components to compensate for the signal deficiency, which was too costly. Sugar’s solution, reached while the engineers pondered which of these two options to pick, was to write-off the fee paid to Ferranti and to get an alternative supplier on the case. He figured that losing that money was as nothing to the loss of face and sales that any delay in the computer’s availability would cause.
Ambit and MEJ switched from Ferranti’s bipolar process to LSI Logic’s CMOS process for their gate array and after no small job adapting the design to the alternative process, got it working. And then Alan Sugar fell out with LSI which forced a second switch, to Italy’s SGS Microelettronica. Fortunately, SGS was second-sourcing LSI’s process so the change was not an onerous one, but Mej had to spend a couple of weeks in Milan making adjustments. Back home, the move from Ferranti’s ULA to LSI’s gate array had already resulted in a change being made to the PCB, which went from version MC0001A to issue MC0002D.
The ULA was home to such features as the 464’s video generator, timing logic - allowing the display controller access to the memory between CPU cycles - the ROM/RAM memory switching, interrupt generation, and the colour palette generator. This latter component output through three pins, for red, green and blue signals respectively. With three possible levels on each - high, low and floating between the two - the 464 was able to deliver its 27 colours.
Alongside the QL
At the CPC 464’s launch in April 1984, Alan Sugar was able to state clearly that all work on the machine had been completed and that its was already being manufactured. He might had said all that anyway, but he was almost certainly encouraged to do so by the highly public delays that Sinclair Research’s latest micro, the QL, was then experiencing post its January launch.
At this stage, Sinclair hadn’t even shipped product and had missed not one but two stated deadlines to do so. When the first QLs began to appear, and no more than a hundred at that, they were found to be buggy and unreliable.
Not so the CPC 464, which, as Sugar had promised, went on sale on 21 June 1984 when a thousand machines were made available to the public by one of Amstrad’s four retail partners, Rumbelows. The quid pro quo for this temporary exclusive – don’t forget, Dixons, Boots and Comet had ordered machines too – was free advertising paid for by the chain. Some 18 software titles were on sale too, from the likes of Gem, Romik, Mikrogen, Bourne, Softspot, Temptation and Britannia.
To be fair to the QL, the Amstrad was using old, established technology - the ageing Z80A processor, clocked to 4MHz; the General Instruments AY-3-8912 three-channel, six-octave sound chip running at 1MHz; a 1MHz Hitachi HD6845 CRT display controller; and an Intel 8255 Programmable Peripheral Interface (PPI) chip. Unlike the silicon in the Sinclair, none of these parts were particularly new. It had one ULA – technically not a ULA since it wasn’t in the end made by Ferranti; it was by now simply a "gate array" – whereas the QL had two, providing much of the Sinclair machine’s functionality.
Alan Sugar said that Amstrad would ship versions of the 464 bundled with either a monochrome screen – green, because Sugar thought it would look more techie – or a colour monitor. The two packages would retail for £239 and £349, respectively, though the colour machine was briefly advertised at £359.
Both machines would feature the long-planned integrated tape recorder, 64KB of RAM and 32KB of ROM for Locomotive’s Basic. The monitor would connect to a Din port on the back of the CPC, which was also home to two edge connectors, one for the parallel printer port, the other for a floppy disk controller. There was a 3.5mm stereo audio output socket, and a d-sub joystick port too.
Sugar announced he would offer a 3-inch disk drive and adaptor, all for £199 and bundled Digital Research’s CP/M 2.2, by then the fast fading standard business micro OS. The games demo’d at the launch were loaded in off a floppy drive hooked up to the back of the demonstration 464.
According to Roland Perry, the decision to use the 3-inch disks had been made long before the launch, after the company evaluated all current floppy formats - 5.25-inch, 3.25-inch, 2-inch and even Sinclair’s Microdrive. That was not considered a runner, and the larger floppy wasn’t right either. The 2-inch format, from Fuji, might have made the cut but for the fact that, unlike the 3-inch drive, it wasn’t connector-compatible with the 5.25-inch floppy drives in the PCs being used for CPC software development.
It wouldn’t be until April 1985 that the 3-inch floppy drive was integrated into the computer itself. The upgraded model was launched as the CPC 664. One problem: there wasn’t enough juice coming through the 464’s 5V DC input routed via the monitor to power the floppy drive. The 664 got around this by incorporating a second, 12V power feed, also via the monitor, which was used to power the floppy drive directly.
Naturally, this required a revised motherboard - the changes saw the joystick and sound ports moved round to the left side of the computer, the addition of a DIN port to allow a standalone cassette recorder to be hooked up to the machine, and the placement of a second edge connector, for generic expansion.
Not that Amstrad needed to worry about the delay - the tape-equipped 464 was, in the summer of 1984, garnering rave reviews and pulling in the punters. “I think the Amstrad will give a lot of sleepless nights to Sinclair, Acorn and Commodore, and nightmares to Memotech and Enterprise/Elan,” wrote Andy Pennell in Popular Computing Weekly, neatly summarising the consensus view. “I think it could turn out to be what the QL could, and should, have been: a terrific home micro with an awful lot of potential as a business machine. Those, myself included, who said that QL was going to be the micro of the year could prove to be rather short-sighted.”
In the Autumn of 1984, Amstrad released the DDI-1 disk interface and FD-1 3-inch floppy drive add-on for the 464. The Interface’s ribbon cable had two connectors to allow two FD-1s to be hooked up as A: and B: Amstrad bundled CP/M 2.2 and the Logo language with it. Again, Locomotive supplied the code: a simple disk operating system that was patched onto the firmware when the drive was connected and the 464 booted up.
Floppy disk generation
This software was built into 1985’s CPC 664 too, and the new machine would also feature an updated ROM containing Locomotive Basic 1.1, which also fixed a handful of bugs, including a duff
DEC$ keyword that had been broken by some over-zealous "byte hunting" as the Locomotive team tried to cram the 464’s firmware into the two 16KB ROM chips. The new micro was priced at £339 for the mono monitor bundle, £449 for the package with the colour screen. It was codenamed "Idiot", short for "Includes Disk Instead Of Tape".
There were some cosmetic changes made too, most obviously the casing colour scheme and the inflation of the cursor keys – on both the 464 and the 664 wrapped around a central, Copy key – into big, MSX-style keys.
The 664 should have been able to run any program written for the 464, but some machine code apps broke when run on the new machine. There developers found out then why Amstrad had requested they make use of its firmware routines through a published table of entry points dropped into Ram when the machine started up. This approached allowed Locomotive to juggle the ROM routines without breaking compatibility with properly written applications and games. Code that bypassed the RAM jump table couldn’t take into account changes made to the ROM by Locomotive. But many of these broken apps were soon updated
The arrival of the 664 prompted some journalists to assume the year-old 464 would now become cheaper. Alan Sugar disabused them of that idea straight away. “I can’t see any reason why the price should change,” he said at the 664’s launch. “Any reduction we may make would be due to reduction in component prices - not through compromising our market.”
Influenced by MSX? Amstrad’s CPC 664
Nor would he supply the 664’s Mk.2 ROMs to 464 owners, or offer any upgrade path from the older machine. Even if he had, users would have been better off waiting a while. Two months later, in June 1984, Amstrad unwrapped a 128KB CPC at the Summer Consumer Electronics Show (CES) in Chicago. News reports from the time are mixed, but it seems that Amstrad said the CPC 6128 was initially destined only for the American market, though it might come to Britain early the following year.
According to Roland Perry, it had been designed at the behest of Amstrad’s US sales partners, who wanted a more capacious machine and one that, after the 464 flopped over there, would look a little more business-like.
Of course, with the cat out of the bag, British resellers began to demand the 128KB model too, and a month after the 6128’s US appearance rumours were flying that the 6128 would debut here early in September, most likely at the Personal Computer World Show. Which is what exactly what happened.
To the annoyance of 664 buyers, the new machine was not only better - updated software and more of it; the extra Ram - but it was cheaper: £229 for the mono bundle, £399 for the colour one. It was a slimmer unit, too, with the Qwerty, cursor and numeric key clusters brought together into a single array. The whole thing look considerably more business like. Meanwhile, gamers after a bargain could now pick up the still extant 464 for £199 or the colour version for £299. With the arrival of the 6128, the 464 finally dropped to the price Alan Sugar had originally wanted to sell it for: just under £200.
“The 6128 is the most attractive new 8-bit micro so far this year. The only people who may complain will be those who have just brought an Amstrad 664,” wrote Jeff Naylor in Popular Computing Weekly. Indeed, it’s telling that Lord Sugar, in his autobiography, skips over the 664 entirely and says the 6128 was the follow-up to the 464. Perhaps all that firing has affected his memory...
Amstrad could get away with it. It was by now better known as a maker of good microcomputers than as a seller of budget hi-fi. In the year to the end of June 1984, into which the then-new 464 had just fallen, computer products accounted for four per cent of Amstrad’s sales. Twelve months on, that proportion had risen to 66 per cent, pushing sale up 60 per cent to £136.1 million of which £20.1 million was profit, itself up 122 per cent.
British home computing: the next generation
And this over a period that had seen Acorn slip off the edge, only to be grabbed and pulled back by Italy’s Olivetti, and for cash crises to send Sinclair Research cap in hand to Robert Maxwell.
Sinclair was spared being forced to sell a large stake to the tycoon for £12m by a big order from Dixons placed with the specific intention of keeping the Spectrum maker afloat. But the writing was on the wall. In February 1986, Amstrad lauded another leap in profits. Sinclair Research plunged even further into the red. It crashed, and Alan Sugar was able to by the Sinclair brand and intellectual property for £5m.
If the pioneers of the 1980s home computer boom were gone or being kept alive with others’ money, Amstrad was on a roll. The CPC series continued to sell nicely, but the September 1985 launch of the PCW 8256 word processor - just a Z80A-based all-in-one with 256KB of RAM, an integrated 3-inch floppy and a bundled printer - sent the company surging forward. Amstrad went on to ship eight million PCW-series machines, more than double the 3m CPCs it sold.
A month after the Sinclair purchase, Amstrad announced its first PC and though this machine suffered several setbacks, the line won good reviews and sales aplenty. In time, Amstrad would ship 12 million of them and become Europe’s leading supplier of PCs.
During this time the CPC line continued to be sold but the shifting nature of the market ensure it wasn’t updated in any way until 1990 when, on the back of an attempt to create a games console to compete with Nintendo and Sega, Amstrad re-styled the line with new, Amiga-like cases. The sound and graphics were souped up, the "CPC" prefix dropped. Like the GX4000 console, the 464 Plus and the 6128 Plus were commercial flops: just too out of date for current gamers’ demands. They were soon dropped from the Amstrad line-up.
Alan Sugar was knighted in 2000 and ennobled in 2009. Throughout the 1990s and the first half of the following decade he was chairman of soccer club Tottenham Hotspur, though he continued to run Amstrad during this time too. The company shifted away from computers, though it acquired computer maker Viglen, and into communications devices and, later, set-top boxes.
In 2007, Sugar sold Amstrad to BSkyB for £125m; Amstrad had by now become Sky’s hardware manufacturer. Sugar himself resigned the chairmanship the following year. He remains chairman of a number of family owned businesses. He has, of course, made regular appearances in TV shows like The Apprentice since 2005.
Roland Perry left Amstrad in 1990. Later that decade he worked for internet service provider UK Online, a move which shifted his focus to the internet and related issues, particularly in the realm of public policy and legislation: since then has been a lobbyist for the Linx, the London Internet Exchange, and handled public affairs for the Internet Crime Forum, the Réseaux IP Européens Network Coordination Centre (RIPE NCC) and now The Internet Policy Agency.
Bill Poel left Amstrad in 1985, became a columnist in the weekly technology trade paper, MicroScope - where this writer had the pleasure of sub-editing his copy - and went on to work in a number of ventures, including Quarterdeck, Enformatica, USP Networks and IncentTV. He’s now UK envangelist for interactive video toolmaker TouchCast.
Marc-Eric Jones continued to develop Amstrad hardware at MEJ, including all of Amstrad’s PCWs, PCs and its portable computers. Having developed a number of EDA tools to help in the design of ASICs and other custom silion during the Amstrad period, MEJ’s business later shifted toward the development and licensing of those tools to chip companies, which prompted a change of name, to 3Soft, and new location: California.
In 1996, Mentor Graphics bought 3Soft, and for the next three years Mej established and ran its IP licensing business. He later steered fabless memory technology company MoSys’ licensing operation. In 2004, he was poached to run Z-RAM memory creator Innovative Silicon. These days he runs Leman Micro Devices, which is working on advanced smartphone sensor technology.
Richard Clayton and Chris Hall continue to work together, though they now run Highwayman Associates, which is now developing technology for software-defined networks (SDN). Locomotive and Amstrad parted ways after the development of the Amstrad PC. In the early 1990s, Locomotive developed Turnpike, an internet access utility, for British ISP Demon; Demon liked it so much it bought the company.
Staying beyond their earn-out periods, the two men quit the company in 2000 and 2003, respectively. Demon was acquired by Sottish Telecom in 1998 and became known as Thus. After leaving Thus, Clayton returned to Cambridge university to pursue his PhD and is now an academic there with a special interest in e-crime. Chris Hall has since taken senior positions at Communications Innovation Institute and cloud services company Interoute. ®
The author would like to thank Roland Perry, Richard Clayton, Mark-Eric Jones, Bob Watkins and Simon Rockman for their invaluable help in the research of this article.