12

I've recently looked into the Sinclair QL schematics and found that it actually has two 8bit ROMs, 16 4164 DRAM chips and yet it is purely 8bit machine, even the 68008 there has 8bit bus.

With very little complication it could be rebuilt into 16bit machine: 68000 instead of 68008, maybe one or two more 74245 or 74373 chips, two or three 74xx glue logic chips and it even could still keep 8bit input into the video ULA (provided another 8 bits are buffered on 74*373). There should be, of course, some more pins for /UDS and /LDS byte strobes, ROM chip selects, etc., but that's basically all. Microdrive ULA could be kept 8bit too.

The PCB itself also does not look overcrowded, so it probably could be fit with 68000 and some extra more 74xx chips

Provided those considerations and that it has ceramic-cased 68008, that was still advertised as 'advance information' in datasheets even in 1985, i.e. it was manufactured in little quantities, if any, and therefore was NOT cheap;

Were there any real cost savings for the QL as a whole from its 8-bitness, or it was just Sinclair's misconception?

  • Have a read of this interesting article on the QL's development. It makes the case for 8-bit not being much of a saving over 16-bit by the end, but that the project would have to be run properly to see it, which it wasn't. rk.nvg.ntnu.no/sinclair/computers/ql/ql_sst.htm – TonyM Jul 10 at 11:09
  • 3
    Good read. I became familiar with crappy Sinclair engineering as a teenager. I used to have 3 Sinclair IC12 modules for a stereo amplifier: 2 were in the amp, and the other 1 was back at Sinclair being replaced (under warranty), on a rotating basis as they took turns dying. I also had friends with the Sinclair Scientific calculator, the Sinclair Black Watch., … cost-reduced beyond their capability. – another-dave Jul 10 at 23:26
  • @another-dave, some familiar kit names there. The magazine adverts for this stuff in the 1968 electronics mags I saw had the same approach as the ZX81/Spectrum ads: a friendly but firm instructive tone about how incredible the gear and Clive were. I inherited a Sinclair Micromatic that was my grandad's. It was amazing in every way: tiny, light, simple...if only it made sound too. – TonyM Jul 13 at 7:29
  • I took a while to learn my lesson - I then bought an Amstrad amplifier. – another-dave Jul 13 at 10:59
  • Ahh, Amstrad audio: hum along to tunes or hum along to the hum, it's all there. The amount of mains hum you got out of their lo-fi gear. First all-in-one with CD player for £200...but I got better sound from my tape deck than their CD player. – TonyM Jul 14 at 14:52
20

Yes, there was a significant cost saving. For a start, Motorola charged us something like 30-40% less for the 68008 (I forget the exact amount, but it was substantial), because the price was so much lower than they were charging their other 68000 customers. Secondly, the pin count issue was key: for the custom chips, going anything over a 40 pin PDIP literally doubled the cost. The alternative would have been to add several HCT245 buffers at around £0.50 each plus more PCB space: for Clive, adding £2 to the cost would have been anathema.

The design would have been much easier and the result much faster. But it would have added perhaps £50-80 to the retail price (remember, retail price is 4x components). So it was out of the question.

David Karlin (I actually designed it!)

| improve this answer | |
  • Gratulation, it's an awesome little machine. Way better than any competition at the time. With real keyboard as a single floppy price might have gone up to 500 GBP, but would have busted all competition and well sold beyond the low end home computer market. – Raffzahn Jul 12 at 12:09
  • This is the best that could happen to questions here -- when the actual designers and engineers explain their decisions. Thanks a lot! :) – lvd Jul 13 at 16:16
  • Would it have been practical to design an 68008 machine to optimize the case of consecutive accesses to two bytes within a word? Perhaps by adding a 74LS373 between the less-significant byte (upper address) DRAM data outputs and the bus, so that when fetching an even byte, the corresponding odd byte would get latched into the latches, and when performing an immediate follow-on address to the corresponding byte, the DRAM could complete its cycle? Or would there have been no nice way to distinguish "other half" accesses from any other kind, or would the timing advantages have been tiny? – supercat Jul 13 at 16:55
  • @lvd Say thank you to peteri on twitter, who sent me the link to this post! – xgretsch Jul 14 at 19:28
  • @supercat Not sure I understand exactly what you're suggesting. It would conceivably have been possible to have a 16 bit memory bus with the 8 bit processor, which would have won a lot of bandwidth - at the expense of PCB space and TTL. It's not something I considered at the time. – xgretsch Jul 14 at 19:29
13

By the time the QL was first designed (starting as a "ZX83" in early 1983), a full-blown 8MHz 68000 was not a mass-produced, cheap commodity item, but rather a pretty expensive beast to buy. Common computers that had it at the time were the Sage and Sun range of workstations, the most common, "mass-market" computer that featured it was the Apple Lisa - Far beyond the QL's targetted market price. (There were others, like the Tandy TRS-80 Model 16, but basically everything that was a full package at that time had a 5-figure USD price tag attached.)

In 1983, an 8-MHz M68000 chip went at a mass-quantity price of roughly 40USD - Way too much for Sinclair's budget. The 68008 was about 1/3 of that. There's a famous quote from Steve Jobs, who went in negotiations with Motorola for the Mac, and was able to cut that going price for the CPU down to $9 apiece, but roughly half a year later when Motorola had optimized the production process, and, of course, with the Apple brand as a backing, rather than largely unknown (in the US) Sinclair.

When Motorola announced the 68008 at a much lower cost, apparently Sinclair saw the chance to earn a bit of the big 68000's fame with the much cheaper CPU and designed for that. Only during the development time of the QL, the 68000 saw a tremendous price reduction on single chip prices (A fact that opened up cost feasibility for the Atari ST, for example) - But then it was too late to re-design (I also pretty much doubt a 16-bit-design would have fitted into the QL's (physical and commercial) package: The Atari ST's PCB is about three times the size of the QL's).

| improve this answer | |
  • 2
    Don't forget the Tandy Model 16, which had sold more than 2000 units by 1983, followed by the 16B selling another 10,000+ units. In terms of units it was the most sold 68k (desktop) system at the time. – Raffzahn Jul 10 at 12:58
  • 2
    @Raffzahn Well, of course - But, being a Muliti-User Unix WS, that was just as far beyond Sinclair's target price range as any other ;) – tofro Jul 10 at 13:07
  • So are the mentioned Sage and Sun, selling in lower numbers. weren't they? The remark was just meant to keep an eye on what sold most, not just what's been famous. I do agree with your view here - especially with the fact that redesigning for 16 bit would have added space issues. – Raffzahn Jul 10 at 13:22
12

TL;DR:

It was all about the incredible low price of 400 GBP. Outclassing any other 16 bit system (except for the TI 99/4) by at least a magnitude, on par or undercutting actual 8 bit machines as well. Every fraction of a penny saved was important.


I've recently looked into the Sinclair QL [...] yet it is purely 8bit machine

I'd say it's an 8 bit system design, but a 16 bit computer.

With very little complication it could be rebuilt into 16bit machine: 68000 instead of 68008, maybe one or two more 74245 or 74373 chips, two or three 74xx glue logic chips and it even could still keep 8bit input into the video ULA (provided another 8 bits are buffered on 74*373). There should be, of course, some more pins for /UDS and /LDS byte strobes, ROM chip selects, etc., but that's basically all.

Doesn't that already sound like a lot? It adds up to several pound. More important, each of this chips does not only raise cost by itself, but as well by the holes to be drilled in every board, thus increasing PCB cost as well.

The PCB itself also does not look overcrowded, so it probably could be fit with 68000 and some extra more 74xx chips

Sure. But it would also have increased design cost. It's a huge difference between routing 8 data lines and doing so for 16. Not to mention additional signals. Professional design is not about cramping in as much as there is space, but as little as possible.

Provided those considerations and that it has ceramic-cased 68008,

Doesn't matter. I bet Sinclair did not require a ceramic package, so it's entirely plausible that it was Motorolas decision to use what they had - at a price point they agreed independently.

that was still advertised as 'advance information' in datasheets even in 1985

Many data sheets carried this label over 20+ years. Saying 'advanced information' or 'preliminary' in case of Intel was one way to avoid costly law suits with mighty customers (*1)

i.e. it was manufactured in little quantities, if any, and therefore was NOT cheap;

That conclusion is not supported by either. It needs concrete proof to make any speculation about price worthwhile. Price is, especially in startup production negotiated independent of production cost, keeping a focus on long term sells and time to market.

Were there any real cost savings for the QL as a whole from its 8-bitness, or it was just Sinclair's misconception?

Of course - like with all Sinclair products every penny counts.

The QL was set out at a very aggressive price point of 400 GBP.

  • That's the same price as a bare bone BBC Model B
  • It was half that the cheapest PC, a SANYO MBC 555 with as well 128 KiB did cost
  • The new Mac was priced at 1500 GBP - almost 4 times the QL. And lets be serious, the Mac can't compete feature wise.
  • A bare bone Apple IIe was 500 GBP, a design of 8 years before.
  • Even a no name Apple II clone was the same 400 GBP
  • Every Portable of the time cost about 400+ GBP(Tandy M100, Olivetti M10 Epson HX-20)

Not to mention that the, at the time most sold 68k desktop system, the Tandy 16/16B, cost about 6000 USD or ~4000 GBP (*3).

Sure, a Spectrum was only about 110 GBP and a C64 ~160 GBP but again, not really a competition here.

Bottom line: The QL was extreme competive priced and every hole, every chip, and every hour in design had to be cut. Going 8 bit external was a good choice.


*1 - Stories that failed projects got ultimate blamed on supplier errors are a household issue. After all, nothing is better than telling management it was some changed or unclear spec from outside - and management tries of course to sue the supplier.

*2 - All prices are taken from a 1984 issue of practical computing.

*3 - Depending on the time of year. 1984 was a bad year for the pound, it lost more than half it's value against the dollar. Which in fact as well hit the QL.

| improve this answer | |
  • In your arguing, you under-estimate very important difference between retail price and components/labour price which basically have nothing to do with each other. – lvd Jul 13 at 16:09
  • @lvd Well, while it is true that price is made to fit the demand side, this only defines the upper ens. The lower end is made up by manufacturing cost. Otherwise I guess we all had golden faucets. For computers a general 4x to 5x number between part cost and sales price is a good rule o thumb. So shaving 10 GBP of of parts results 40+ GBP possible lower sales price. And with a goal st to undercut everyone, as so often with Sinclair products, Margins are cut, so we operate even closer to that than anywere else. – Raffzahn Jul 13 at 19:35
4

In complement to the other answers, never forget that in the 1980s memory was expensive.

For a 16-bit system, the base memory and every expansion costs twice the price of an 8-bit system - perhaps £100 more at the time.

This is a big reason why Intel sold so many 8088s (there were other cost savings in bus buffers and PCB).

Motorola tried to follow suit with the 68008, but just from looking at the instruction set, it was an unwise move. The 68000 instruction set is based on a 16-bit word, whereas the 8086 instruction set is based on the byte.

On a 68008, every instruction requires two fetch cycles, whereas the 8088 has a number of 1-byte instructions.

MOVE.B #0, R1 on the 68008 requires four fetch cycles, whereas the equivalent MOV AL,0 on the 8088 requires two fetch cycles.

So an 8-bit QL was cheaper on memory, and Sinclair probably got a good price from Motorola on the 68008 because it was too slow to be a candidate for a serious system.

Confession: I persuaded my father to buy a QL because I would never have persuaded him to buy the applications he needed, but also so I could learn about the 68000.

| improve this answer | |
  • 3
    The only time memory price would be affected would be if an odd number of 8-bit wide banks would have a usable amount of memory. If one would need an even number of 8-bit wide banks, using a 16-bit bus would require doubling up a small amount of memory-select logic, but the amount of extra circuitry would be relatively slight compared to the cost of the memory. – supercat Jul 10 at 23:01
  • @supercat in the 1980s, memory was installed by chip, either 1 bit wide or 4 bits wide. It was also slow (250 or 300ns), so doubling the banks as you suggest would negate the advantage of a 16-bit external bus. – grahamj42 Jul 11 at 7:13
  • 1
    My point was that if one wanted e.g. 32K or 64K in the era of 16Kbit chips, or 128K or 256K in the era of 64Kbit chips, one would need 16 or 32 RAM chips, regardless of whether they were wired as two or four rows of 8, or as one or two rows of 16. When 64Kx1 RAM chips were available, using eight of those might have been cheaper than thirty-two 16Kx1 chips, but unless all RAM would fit in eight Nx1 or two Nx4 chips, the cost of RAM would be unaffected by an 8- or 16-bit bus width. – supercat Jul 11 at 13:03
  • 1
    The QL already has two 27128 8-bit ROMs and sixteen 1bit 4164 DRAMs, so they could be combined as 16-bit memory as well. This observatios was the basis for my question. – lvd Jul 13 at 15:58
  • 1
    "This is a big reason why Intel sold so many 8088s (there were other cost savings in bus buffers and PCB)." - the main reason the 8088 sold so well is that intel offered it extremely cheap to IBM, and then clone manufacturers felt they had to follow suit in order to be properly compatible... – occipita Jul 13 at 22:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.