On 15/06/2017 05:40, David Froble wrote:
> Arne Vajhøj wrote:
>> On 6/14/2017 6:05 PM, David Froble wrote:
>>> Paul Sture wrote:
>>>> On 2017-06-14, David Froble <***@tsoft-inc.com> wrote:
>>>>> Alpha is now over 25 years old. It was planned to be usable for those
>>>>> 25 years. Got to consider it successful, from that perspective.
>>>>>
>>>>> The question I have is, which semiconductor company will be designing
>>>>> the next "new" processor? When there was competition, such was
>>>>> happening. It's sort of depressing to think of x86 as being an
>>>>> eternal design ....
>>>>
>>>> ARM seems to be the competition for the future.
>>
>>>> "With over 100 billion ARM processors produced as of 2017, ARM is the
>>>> most widely used instruction set architecture in terms of quantity
>>>> produced."
>>
>>> Now, I'm just a dumb polock, come down out of the hills, but it's my
>>> impression that ARM targets the low end commodity market. Could be
>>> wrong.
>>>
>>> What I was referring to is the same targets Alpha was aimed at, ie;
>>> super performance, the CPU of "the future".
>>
>> You are absolutely right ARM is targeting phones and other low end
>> commodity markets.
>>
>> But let us go back 20-25 years. What was status back then?
>>
>> SPARC, Power, Alpha etc. was targeting the super performance
>> market.
>>
>> x86 (later evolving into x86-64) was targeting low end
>> commodity markets (PC's).
>
> At the time the volume was in desktop PCs. Could they do everything?
> No, but, they sure could be cheap.
Commodity PCs could appear cheap to buy. Business-class PCs with a
lifecycle of more than a couple of months cost a bit more.
>
> DEC thought they had to get Alpha into PCs, or something like that. Did
> they really? Not so sure. But, their choices, and the changing
> computing environment, and lack of vision, and other things, did DEC in.
Did you try doing a price comparison between volume orders for
business-class desktop PCs and a comparable desktop Alpha system,
e.g. in the late 1990s? I did.
E.g. in the Personal Workstation range, based on industry standard
NLX-format boards, with the processor on a daughtercard, you could
have either an x86 system or an Alpha EV5 system, with
surprisingly small price difference (because everything except the
processor card was the same), and (for certain applications)
there was a very worthwhile performance benefit on the Alpha.
https://en.wikipedia.org/wiki/Digital_Personal_Workstation
One such application was converting PostScript to bitmap for
professional printshops (then known as Raster Image Processing).
Performance and reliability were key, and NT/Alpha was a good fit
especially with the PWS range (which, for those willing to pay
the price premium for the same basic hardware, would also run VMS
or Tru64).
Then MS and Intel effectively pulled the plug on NT/Alpha and
suddenly there wasn't much future for NT/Alpha anywhere. And that
and a few other Intel/MS-related management issues meant that
there wasn't much of a future for Alpha anywhere.
>
>> Did SPARC/Power/Alpha kill x86 or did x86 kill SPARC/Power/Alpha?
>
> IBM took a different course, and Power is still with us, and is top of
> the line. It will never have the volume x86 had at one time. I don't
> think x86 on the desktop still has the market it once had. Not too many
> XEONs and Opterons in desktops.
>
> I don't feel that it's the CPU designs, nearly as much as the companies
> making them. Look at Intel. Look at their dedication to the itanic.
Intel were only dedicated to the Itanic till AMD64 came along and
showed that, contrary to Intel's public claims, x86-64 wasn't
impossible. Once AMD64 made x86-64 go public, and AMD64 scalability
showed that it was ready for what most people called "enterprise
class" systems, the limited future of IA64 was clear.
> Then look at IBM, knowing they need a better CPU to differentiate
> themselves, and they have the dedication to stay the course. If you
> want to talk about "killing", talk about management, not design.
>
>> (well SPARC and Power is not dead yet but close to gettting
>> into intensive care)
>
> Some people have been saying that for a long time. Don't know about
> Sparc, but I'm thinking Power will be around for some time yet.
>
>> ARM is right where they should be to be serious competitor!
>
> In what market?
ARM has been in (and often dominated) most volume markets where
Windows/x86 isn't a requirement. For most of those markets,
customers (who are not end users) can get almost everything the
system needs, all on one chip (processor, memory, IO, etc) -
this is what the industry calls 'system on chip', aka SoC.
ARM works everywhere, from the electronics on your disk drive, to
your smart TV, to your SoHo network gear. Oh, and mobile phones
and tablets and almost every other piece of kit with a computer
inside, so long as it doesn't need to run Wintel software native.
Legacy x86 designs are reliant on revenue from volume desktop
and notebook chip designs (which largely don't sell anywhere
else) to make lower-volume high end x86 chips affordable. As
the relevance of desktop and notebook designs decreases, and
the non-recurring costs of high end x86 designs increase,
the end user cost per high end x86 CPU will necessarily
increase - just as it did for Alpha.
Intel still haven't succeeded in doing SoC in volume, not
least because Intel can't do cheap x86. If they do cheap
x86 with reasonable performance they're just undercutting
their higher-end prices anwyay. For a while they tried to
hide the pricing impact of this by subsidising their SoC
products, but once it became obvious, they had to re-organise
the company accounts to keep it hidden. Look for 'contra
revenue' articles, e.g.
https://www.cnet.com/uk/news/behind-in-tablets-intel-pays-firms-to-use-its-chips/
In addition to those pricing issues, it's recently emerged
that some of Intel's SoC designs (specifically, Intel Atom
C2000 products based on new-to-Intel Finfet fabrication
technology) in routers, NAS boxes, etc, have had life-limiting
reliability issues after a few months, and it's been hushed
up:
https://www.theregister.co.uk/2017/02/07/intel_atom_failures_go_back_18_months/
Intel are not done yet though, especially given last year's
unforeseen purchase of ARM by Japan's Softbank, and earlier
this year there was talk of Softbank selling a quarter of
ARM to Middle Eastern investors...