RISCOS 35th Birthday

on-topic acorn-related discussions not covered by the other forums
Post Reply
User avatar
a1exh
Posts: 147
Joined: Fri Apr 16, 2021 5:35 pm
Location: Oxfordshire
Contact:

RISCOS 35th Birthday

Post by a1exh »

I just noticed this post on TheRegister

https://www.theregister.com/2022/06/21/risc_os_35/
Principal ASIC Engineer
RetroGeek
Acorn Electron + A5000
http://thalion.atari.org
paulb
Posts: 1101
Joined: Mon Jan 20, 2014 9:02 pm
Contact:

Re: RISCOS 35th Birthday

Post by paulb »

a1exh wrote:
Wed Jun 22, 2022 5:20 pm
I just noticed this post on TheRegister

https://www.theregister.com/2022/06/21/risc_os_35/
A correction:
RISC OS has had a rather convoluted history, partly due to Acorn spinning out Arm, eventually pulling out of the computer market, rebranding as Element 14 and being acquired by Broadcom, where Arm co-designer Sophie Wilson still works today.
In fact, although Acorn rebranded as Element 14, the company of that name acquired by Broadcom was a separate company set up by a certain Acorn executive to acquire various Acorn assets. Meanwhile, Acorn Group and Acorn Computers Limited were renamed and continued to exist as holding companies, these having been set up by the Morgan Stanley subsidiary who acquired Acorn. The former Acorn Computers was apparently parked in the Cayman Islands (see Cabot 2 in this list): an indication of the financial acrobatics at work.

Forbes was more charitable, referencing a "buyout offer from its management" in the context of the Element 14 branded version of Acorn Computers, but apart from the corporate discontinuity (Cabot 2 actually administering things like the rights to various Acorn assets, not the "buyout" company, nor Broadcom subsequently), the "new" Element 14 was effectively an airlift of the "silicon design" activities out of Acorn that had been set up in the year or so prior to Acorn being dismantled. As everyone knows, the parts of Acorn that had generated the revenue over the years were put out to pasture, leaving customers, developers and dealers in the lurch.

Personally, I would be more interested if someone could actually find anyone who worked on ARX (or RISC iX for that matter) and heard their stories.
User avatar
iomanoid
Posts: 670
Joined: Sat Aug 08, 2009 10:38 am
Location: Baseworld: Cygni
Contact:

Re: RISCOS 35th Birthday

Post by iomanoid »

Who owns Acorn Computers Ltd now, do you know?
User avatar
BigEd
Posts: 4944
Joined: Sun Jan 24, 2010 10:24 am
Location: West Country
Contact:

Re: RISCOS 35th Birthday

Post by BigEd »

Companies have names, which can change, but also have numbers, which don't - as I understand it, the company formerly known as Acorn Computers Ltd became the company known (perhaps briefly) as Cabot 2 Limited. (Yes, see here.) And when companies are dissolved, or perhaps merely renamed, the name can be reused. While the company formerly known as Acorn Computers Limited was for a short while known as Element 14 Limited, the "airlifted" new company, initially known as NewJam, was renamed to Element 14, but I think in both cases not UK Limited, but US Incorporated.

When the assets and employees and business relationships get transplanted from one company to another, and the company gets renamed, it can be difficult to keep track.

But it's true that Sophie still works at Broadcom.

And the company that's presently named Acorn Computers Limited is owned by someone not very far from here... but it's a completely unrelated company.
paulb
Posts: 1101
Joined: Mon Jan 20, 2014 9:02 pm
Contact:

Re: RISCOS 35th Birthday

Post by paulb »

BigEd wrote:
Thu Jun 23, 2022 7:13 am
Companies have names, which can change, but also have numbers, which don't - as I understand it, the company formerly known as Acorn Computers Ltd became the company known (perhaps briefly) as Cabot 2 Limited. (Yes, see here.) And when companies are dissolved, or perhaps merely renamed, the name can be reused. While the company formerly known as Acorn Computers Limited was for a short while known as Element 14 Limited, the "airlifted" new company, initially known as NewJam, was renamed to Element 14, but I think in both cases not UK Limited, but US Incorporated.

When the assets and employees and business relationships get transplanted from one company to another, and the company gets renamed, it can be difficult to keep track.

But it's true that Sophie still works at Broadcom.
All I was really objecting to was the fiction that there is some kind of corporate continuity between Acorn and Broadcom, which may suit someone's narrative where Acorn transformed itself and was gloriously purchased by a huge corporation, but the fact is that Acorn itself was beached in the Cayman Islands and, after a bit of negotiation with the board, a bunch of people went off and did their own thing with some of the company's assets. If one wanted to claim any kind of vague institutional continuity, one could more credibly argue that all the people who ended up at Pace were the continuation of Acorn in some sense closer to what Acorn were traditionally doing. There were other companies who kept up the various activities of Acorn after its demise, too.

I suppose one could portray any company involving Sophie Wilson, Andy Hopper, Hermann Hauser and a sack of investment capital as being "Acorn", but I think it would be rather insensitive to everyone else who worked there. And perhaps the thing of most concern to those of us interested in the history is the matter of the custodianship of the institutional knowledge and official history of the company. If Broadcom had indeed inherited the Acorn mantle, they might have kept hold of such resources and archives, not least because such companies love to litigate on occasion if they feel they have some basis for it, whereas it seems to be the case that all that stuff was dumped in a skip at some point in 1999.
BigEd wrote:
Thu Jun 23, 2022 7:13 am
And the company that's presently named Acorn Computers Limited is owned by someone not very far from here... but it's a completely unrelated company.
Yes, I was discussing that acquisition with the proprietor a few weeks or months ago, having noticed it probably rather later than everyone else. I think that given his record in preserving materials from the era, I have far more confidence in whatever plans he might have for it, if any, than of those who have had the nerve to claim any kind of continuity - institutional or cultural - with a company that was effectively discarded when a convenient opportunity arose to make some nice money.

Then again, there are people who profess to care about the fate of ARM (or Arm, or whatever), lobbying for UK government intervention and lamenting the potential consequences of a sale of the company by its current owner, even though the same people happily took the money when that same company acquired the business and presumably made such people very wealthy (or even wealthier).
User avatar
BigEd
Posts: 4944
Joined: Sun Jan 24, 2010 10:24 am
Location: West Country
Contact:

Re: RISCOS 35th Birthday

Post by BigEd »

Very good point, trying to draw continuity with Broadcom is misleading (or a mistake!) (And yes, if there were any validity, one would also want to connect Pace, and possibly others.)
paulb
Posts: 1101
Joined: Mon Jan 20, 2014 9:02 pm
Contact:

Re: RISCOS 35th Birthday

Post by paulb »

Another thing that the article touches upon, by referencing the SPARCstation 1 and DECstation 3100 as being relatively costly workstations available only in 1989, is perhaps the notion that the Archimedes was ahead of the curve as a RISC-based machine. Certainly, in terms of price, there weren't exactly that many low-cost RISC systems around in 1987, but in terms of RISC machines generally, the Archimedes was notably preceded by the IBM RT PC (ROMP) system which was launched in 1986. Admittedly, the RT PC cost over $16000, but it did run Unix:

"IBM unveils RISC system", Computerworld, 27 January 1986.

Interestingly, IBM's RISC offering and the ARM were categorised similarly, at least by one competitor:

"What's All The Fuss About RISC", UNIX Review, February 1986.

In that piece, the author (affiliated to MIPS Computer Systems) makes the following remarks:
Figure 2 attempts to put the RISC movement in context by showing performance projections for anticipated commercial VLSI RISC processors. Two RISC performance groups are shown. In the lower of the two we find chips in the 1.5-3 MIPS class; this includes the sort of chips that might be designed for single-user personal computers or controllers. The aim in this range is to provide good performance at a low cost, often with cacheless operation. Among the examples are the ACORN RISC (announced in 1985) and IBM's new entry, the RT PC (running at 1.6-2.1 MIPS).
The figure effectively shows the lower-performance RISC chips having a progression mostly corresponding to Motorola's 68000 series roadmap, whereas the upper-performance chips have a progression with a steeper gradient mostly corresponding to Digital's VAX roadmap. It also notes that at that time, only the Fairchild Clipper had been announced, although MIPS themselves entered the market in 1986 with the R2000 and accompanying chipset. It should be noted that when the author refers to "MIPS" as a performance measure, he is referring to the VAX 11/780-equivalent performance unit. The Archimedes benchmark result in the Byte article referenced by the Register appears to give that system a performance rating of around 2.8 VAX MIPS, so at the upper end of that lower-performance group.

Back in 1986, when the IBM RT PC came out, it appears that observers were more interested in what HP were going to do, and HP announced their "Spectrum" (PA-RISC) products in March 1986:

"HP launches long-awaited RISC effort", Computerworld, 3 March 1986.

As has been noted in various talks, HP took a while delivering products, doing so in September 1987:

"Long wait ends as HP ships top-line Spectrum", Computerworld, 28 September 1987.

Meanwhile, the first independent MIPS-based workstations (as opposed to the development systems available for vendors) seem to have been the Silicon Graphics Iris 4D and Prime Computer PXCL 5500 models, these having a somewhat related development heritage, arriving in early to mid-1987. They are far from cheap machines, as one might expect with 16MB of RAM and being targeted at high-performance markets:

"Powerful chips threaten departmental systems", Computerworld, 6 May 1987.
"Slew of intros marks Siggraph", Computerworld, 27 July 1987.

Certainly, with "over 30 custom and semi-custom graphics processors", the PXCL 5500 was aiming rather higher than things like the DECstations, let alone Acorn's products.

What isn't really said in the Register article is that workstations were generally assumed to be running operating systems such as Unix, or at least delivering many similar capabilities, and there was going to be a need for rather expensive amounts of RAM to make that feasible. It has been noted before that Acorn repeatedly pitched workstation products at around £4000 or so, with the Cambridge Workstation costing that but arguably not even qualifying as a workstation by that time, but one has to wonder if they weren't aiming too low with the ARM-based machines.

The Archimedes as it was originally delivered has some workstation-like characteristics (such as a graphical user interface), but it didn't really scale up to compete with other workstations successfully. I find it insightful that the UNIX Review article above emphasises the cacheless nature of the ARM, suggesting that the author understood the design philosophy fairly well, acknowledging that it suited cheaper systems. But it is also well-known that the MIPS architecture emphasised the use of cache memory from the beginning, whereas it took a few years for the ARM3 to come along with a cache.

Another observation that perhaps runs in parallel with the scalability of the hardware, or lack thereof, was the operating system strategy at Acorn. Just as the ARM eventually struggled to keep up with the performance advances more generally, RISC OS was never really going to stretch to handle the workstation-like sophistication of personal computing in the 1990s. But anyway, I think I've gone on enough!
User avatar
BigEd
Posts: 4944
Joined: Sun Jan 24, 2010 10:24 am
Location: West Country
Contact:

Re: RISCOS 35th Birthday

Post by BigEd »

paulb wrote:
Sat Jun 25, 2022 4:11 pm
I find it insightful that the UNIX Review article above emphasises the cacheless nature of the ARM, suggesting that the author understood the design philosophy fairly well, acknowledging that it suited cheaper systems. But it is also well-known that the MIPS architecture emphasised the use of cache memory from the beginning, whereas it took a few years for the ARM3 to come along with a cache.
One observation of Steve Furber's, I think, is that the transistors used for cache in a RISC are in some sense playing the same role as the transistors used for microcode in a CISC. I wonder when it was that Steve first picked up on this idea, and I wonder how early in the ARM design it was clear that a later chip would be able to include more transistors and include a cache. In his 1989 book VLSI Risc Architecture and Organization he writes
A feature of microcode is that the microcode ROM usually has very good access time, so frequently used operations will run fast if they are microcoded. One of the earliest RISC exponents (Radin, 1983) pointed out that cache memories also have good access times, and whereas microcode only contains the static set of operations chosen by the original designer, a cache can contain a dynamic set of frequently used operations selected automatically by the hardware to suit the current task.
Maybe see also this Furber observation:
So we did continue with chip developments. We developed ARM3, which was effectively putting ARM2 onto – I think it was a one or one and a half micron process, which left enough space to put a cache around it. So that was the first ARM processor that had a cache memory, so it stores frequently used things locally on chip. And that was hugely advantageous for the Archimedes because the – although in principle the cache allowed us to push the clock up to twenty megahertz and go two and a half times as fast, in practice the Archimedes processor was sharing memory bandwidth with the graphics. The cache reduced its bandwidth requirements, so in practice Archimedes machines with ARM3 went about four or five times as fast as the ARM2 based machines. So we got a huge win out of ARM3.
Post Reply

Return to “general”