Re: [?] Why should Distros be called as i386 for a 32-bit PC, and as amd64 for a 64-bit PC, when Intel Core PCs are also 64bit systems
On Sun, Mar 14, 2021 at 10:44:00AM -0500, John Hasler wrote:
The Wanderer wrote:
It caught on, and became so successful that Intel abandoned its ia64
approach and started making amd64 CPUs itself.
Which was unfortunate as the x86 architecture needed to die.
Moving to ia64 would have been much, much worse. Luckily it was unlikely
to have ever happened once people got to touch actual silicon.
On Sun, Mar 14, 2021 at 02:50:10PM -0400, Stefan Monnier wrote:
So it was a great move on the part of AMD: cheap to implement but with
an enormous marketing impact.
It had much more than a marketing impact, because x86 was a PITA for
more than 2GB of RAM and that was getting cheap and becoming a common
problem by 2003. Switching to opteron for 8G or 16G servers was a huge
win vs x86, with better scaling for multiprocessor configurations.
(These were becoming more common as well, and intel was still using an
old (obsolete?) flat SMP bus whereas AMD arrived on the scene with a far
superior NUMA architecture based on hypertransport--designed in
partnership with what was left of the old DEC alpha team.) It was simply
the right product at the right time.
On Sun, Mar 14, 2021 at 03:17:39PM -0400, Stefan Monnier wrote:
But years passed and the i386 architecture got better and better,
including stuff like MMX, SSE and AVX was incorporated, IA64 couldn't
really keep up.
The IA64 architecture was a resounding success in one area tho: it
killed most of the competition that was coming from "above" (at least
DEC's Alpha, SGI's MIPS, HP's PA, and it likely sped up the demise of
Sun's SPARC, I don't think it had much impact on POWER or PowerPC, OTOH)
and thus helped open up the server (and supercomputer) market for Intel
(and AMD).
Yes--SGI, HP, & DEC (Compaq then HP) all preemptively killed off their
CPU lines based on the promises made for ia64. When ia64 turned out to
be late and the performance turned out to be disappointing, it was too
late to revive their previous architectures and recapture the customers
that had already abandoned ship for x86 & later amd64. It worked out
really well for intel, and really badly for everybody else.
On Mon, Mar 15, 2021 at 09:15:10AM +0100, Sven Hartge wrote:
Stefan Monnier <monnier@iro.umontreal.ca> wrote:
The IA64 architecture was a resounding success in one area tho: it
killed most of the competition that was coming from "above" (at least
DEC's Alpha, SGI's MIPS, HP's PA, and it likely sped up the demise of
Sun's SPARC, I don't think it had much impact on POWER or PowerPC,
OTOH) and thus helped open up the server (and supercomputer) market
for Intel (and AMD).
I think, IBM is big enough and old enough and established enough with
POWER that a "young whippersnapper" like Intel is no real danger to them
in their own enclosed Mainframe walled garden. I believe Apple moving
away from PowerPC did more damage to IBMs aspirations in that market.
IBM didn't want to just be a mainframe manufacturer, they really wanted
to amortize the costs for those CPUs against multiple product lines.
They actually made a good number of high end computing sales for a few
years by being the only player left standing, until amd64 just became
too compelling. They still have some very large deployments, but their
overall market share is not what they'd hoped for.
For the others: they where either on board from the start (like HP),
where already dead (like DEC/Compaq) or slipping into the embedded
market (like MIPS).
At the time ia64 was announced alpha & MIPS processors were in some of
the largest and most sucessful systems in the world. With further
development they could have remained there, but their management was
convinced that ia64 was going to have an unbeatable performance
advantage and that they couldn't compete with the R&D money intel was
pouring in. With hindsight it's clear that neither was true but these
decisions were made in the late 90s and intel hadn't yet run into the
brick wall of making the compiler magic actually work. The architecture
that was in the worst shape was PA-RISC--which is why HP had gone in
with Intel on ia-64 in the first place. (And, of course, the alpha had
no future once HP bought Compaq.) Also with hindsight, even if ia64 had
been successful this strategy would have destroyed the companies because
it was premised on the idea that even if they were all selling the same
computers they'd somehow be able to keep their margins and lock
customers in with proprietary OSs or some other proprietary magic. The
industry went in a very different direction and preferred open software
architectures, and that probably would have been true even with a
successful ia64. HPaq & SGI bet on the wrong horse in every way.
The cloud revolution of the 2010s might have unfolded very differently
if some of the high performance architectures from the late 90s could
have hung on long enough for the linux convergence to offer them a way
out of the unix wars. (Or, they might have kept the unix wars going. Who
knows.) Instead, people are only now trying to break out of the
monoculture by pushing what was 25 years ago one of the least successful
and least capable of the RISC architectures (ARM) into the
high-performance realm, for lack of other options in a space utterly
dominated by amd64.
Reply to: