[Info-vax] Linux support

[David Froble]

johnwallace4 at yahoo.co.uk wrote:
> On Friday, 31 January 2014 18:48:05 UTC, David Froble  wrote:
>> johnwallace4 at yahoo.co.uk wrote:
>>
>>
>>
>>> There's another discussion to have about what will happen to Intel's x86 
>>> pricing in a year or five, now Intel can clearly no longer rely on historic 
>>> x86 volumes in the client computing (and low end server) upgrade treadmill 
>>> market. SPARC and IBM won't have that 'problem' because they don't have that
>>> volume. But although that's a related topic, it's separate, and possibly 
>>> further out.
>>
>>
>> You could find some answers by asking why Intel has a 64 bit x86 product 
>>
>> line.
>>
>>
>>
>> Simple answer, AMD.
>>
>>
>>
>> Now, possibly you've been lumping Intel and AMD together for your above 
>>
>> statements.  Doesn't seem to work that way.
>>
>>
>>
>> While it appears that Intel currently has the performance advantage, 
>>
>> it's not like it's a huge difference.  When it comes to price vs 
>>
>> performance, AMD seems to be ahead.  Need a bit more, add a few more 
>>
>> cores onto the chip.  I've got to think that in most cases, absolute 
>>
>> performance is not a big issue.
>>
>>
>>
>> Why didn't Alpha survive.  Perhaps because lesser chips were "good 
>>
>> enough", and the same could be said for the top of the line x86 server 
>>
>> chips.  Throw enough of the lower priced x86 chips at a problem, and 
>>
>> most of the time it's "good enough".  Not always, but for those few 
>>
>> cases, will they support the development and production of high end 
>>
>> chips?  Didn't for Alpha.  Seem to for Power.
>>
>>
>>
>> My opinion is, as long as there is direct competition, the prices will 
>>
>> stay low.  Long live AMD ....
>>
>>
>>
>> You might say that the desktop market will also shrink.  Well, all most 
>>
>> ever wanted was a smart phone or a tablet, so I'd agree that the desktop 
>>
>> market will shrink.  Perhaps the cost will go up a bit to make up for 
>>
>> the lower volume.
>>
>>
>>
>> Perhaps an argument could be made in favor of IBM.  They seem to have 
>>
>> what the top end customers seem to want, good solid hardware, good 
>>
>> support, and good customer service.  The few customers who feel they 
>>
>> need such may gravitate to IBM.
>>
>>
>>
>> Interesting times ....
> 
> Not disagreeing in general but want to add a couple of comments:
> 
> "Need a bit more [performance], add a few more cores onto the chip." 
> 
> Works great for marketing, how well does it work for typical volume-market
> single-threaded workloads on desktops or mobile devices? Readers here know
> the answer...

Everyone stopped making single core chips years ago.  I once made the 
comment that AMD's San Deigo chip at 90 nm was basically the fastest 
single core CPU that would be manufactured.  Got jumped on by a few 
people, but, this was AFTER both Intel and AMD stated that all future 
production would be multi-core.  You may purchase a "single core" chip, 
but that's because for mfg or marketing reasons some of the cores are 
not operational.  So, I stand by my claim.

You're 100% correct, some things are single threaded, and the only way 
to improve performance is to run faster, use cache and faster memory, 
and such.  But the technology has run up to and into some physical 
barriers.  Process shrinks are great, until you don't have enough 
insulation between paths.  The electric isn't going to exceed light speed.

And so, the manufacturers have used the process shrinks to, among other 
things, put more cores on a chip so they can claim greater capabilities.

> "Why didn't Alpha survive.  Perhaps because lesser chips were "good enough"
> 
> By the start of 1990s, available and visible x86 chips weren't good enough
> to last through the 1990s, and nor was PA-RISC or VAX. Intel and AMD both
> knew it, as did DEC and HP. They were all working on 64bit stuff (as was the
> rest of the market - there's a Byte from 1993 that has all the main 64bit 
> players on the front cover, if I remember rightly: DEC, IBM, MIPS, Sun but
> maybe not AMD and not Intel?). 1993 - the year NT 3.1 came out retail?
> 
> DEC HQ (Palmer etc) allowed themselves to be fooled by Intel's promises re 
> IA64. Combined with MS quickly abandoning multi-platform NT, Alpha didn't 
> really have much chance, though it did quite nicely considering the lack
> of HQ support.

Intel and Microsoft are rather sleezy operations ....

> That's a short version of a much longer story.

Yeah ....

> "as long as there is direct competition, the prices will stay low.  Long live AMD".
> 
> Yes long live AMD. But... Intel's success at the higher performance x86 end is 
> dependent on both chip architecture improvements and (historically) on chip 
> fabrication improvements. If they can't fill the chip factories by selling x86 
> in volume, their per-chip manufacturing costs go up (which means retail prices 
> go up or per-chip profits come down). If you can't fill the fab you also make 
> it harder to pay for the next round of chip factory upgrades. If your volumes 
> decrease your cost-per-chip of the non recurring engineering costs also goes up 
> (same problem Alpha had). Slippery slope?

You really have to ask, how much smaller can they go?  If not much, then 
advances have to come from elsewhere.  From what I've seen, most 
advances came from process shrinks, allowing more cores, and more on 
chip cache, and faster speeds.

With a significant reduction in demand, is it possible that advances 
will slow down?  That would mean using the existing fabs for longer 
periods of time.  Fabs might be similar to factories making DVD drives, 
once set up, production is cheap.

It appears that the fabs also seem to make improvements within a 
particular size.  Always learning.  But the improvements are in yield 
and very small if any speed.  Possibly fixing errors.  Nothing spectacular.

One development is the inclusion of graphical cores on the CPU chips, 
and the use of those cores for some operations.

> Are AMD still significant users of other people's chip factories ? Do they 
> need absolutely leading edge fab technology or does "good enough" suit AMD?
> 
> "Perhaps an argument could be made in favor of IBM."
>  
> Quite. IBM don't seem to be greatly bothered by continuing with IBM-proprietary 
> stuff, be it chips, systems, software, or networking. Many of their customers 
> don't seem that bothered either.
> 
> HP are not IBM (and nor were Compaq or DEC, though there may have been times
> when their respective HQs thought they could be).

Speculation.

If everything had happened at once, desktops, notebooks, tablets, and 
smart phones, might CPUs such as Alpha have had a better chance?  There 
might not have been the all invasive x86 to bury most others in it's 
economies of scale.  The 2 things that made x86 was that it was used 
everywhere, and AMD's 64 bit extensions.  Perhaps x86 might not have 
been able to bury Alpha and Power and such because it would not have had 
the huge numbers.  Keep in mind, Intel and AMD have spent big bucks to 
wring performance out of the architechure.  Without cheap x86, the 
customers who still needed the "mainframes" (for lack of a better label) 
would have continued to pay more, and the development would have 
continued to be feasible.

/Speculation