Sad day for Intel and America

an era in US tech

s ago, Intel held out, a strategy that is now in tatters, with the company ?s factories struggling to keep up."

rnings

t screwed

ile

has

tel

nm

nm.

tion

is is

just

uld

ely

ht

Intel.

IF true, then it's odd that so much of the internet and modern computing runs so successfully on it.

Not internal politics, lack of customer acceptance. None of those were ver y good. The Titanium was pretty much targeted at one customer, HP, it was way late. The 960 had stiff competition and didn't gain enough design wins to establish it as an architecture and it was targeted at markets where the X86 could not compete. It's not like the X86 team had to come with axes to kill it, they didn't even pay any attention to it.

Not so much odd as sad. IBM picked the wrong cpu for the PC.

ote:

Huawei.

picking consensus from other government officials. The decision to ditch H uawei come from many branches of the government, representing the people's will.

nobody anything.

the Chinese sea

etnam.

President.

ed just to pester Huawei.

Starting from Mo's era, CCP has been following Sun Tzu's "The Art of War". Which means lies and deceptions to take advantages of others at all cost.

thers.

to try to convert the rest of the world to their brand of Islam, but they d on't seem to be interested in destroying anybody for destruction's sake. It 's more a matter of discouraging other people from trying to stop them doin g their thing.

Like ISIS? They didn't declare war explicitly, but the actions speak of it self.

It's pretty fast _now_ in part because about a billion man-hours over several decades were dumped into writing very good compilers for an odd duck of an architecture.

Imagine where mainstream computing could be at if the same man-hours had been put into writing very good compilers for an inherently more efficient and less compromised architecture.

You can build a racing snail but there's an opportunity cost to doing this.

The biggest markets for CPUs are cell phones and embedded gadgets, and they don't use X86. Intel blew it in both. The big market for crunching is in server farms, and they are drifting away from x86.

Computing is just too cheap nowadays to support Intel's premium-price business model.

Intel bought Altera to defend its server market, to add hardware boost to the clumsy x86. I wonder how that is working out.

Sounds complex.

uawei.

picking consensus from other government officials. The decision to ditch Huawei come from many branches of the government, representing the people's will.

nobody anything.

the Chinese sea

tnam.

resident.

d just to pester Huawei.

Starting from Mo's era, CCP has been following Sun Tzu's "The Art of War". Which means lies and deceptions to take advantages of others at all cost.

hers.

not to use them again.

You mean when we had all the nukes?

Though forward-thinking in many aspects, the Baruch Plan had several drawba cks. The most controversial one was probably its insistence that the United States retain its nuclear stockpile (which then consisted of nine weapons) until it was satisfied with the effectiveness of the international control system.

I can't imagine why other powers would feel at all threatened by that.

f an era in US tech

rs ago, Intel held out, a strategy that is now in tatters, with the company ?s factories struggling to keep up."

arnings

st screwed

hile

has

ntel

ction

his is

s.

s just

ould

kely

ght

Intel.

ut

nt. Wow. And IDK how you make something inherently elegant, whatever that even

Current x86 processors don't "maintain" the x86 architecture. They are in ternally a RISC design and emulate the x86 instruction set via microcode.

Words like "elegant" and "kludgy" have no engineering value. The only thin g that is important are the measurable parameters.

era in US tech

o, Intel held out, a strategy that is now in tatters, with the company? ??s factories struggling to keep up."

gs

rewed

l.

So what makes TSMC so special? It's not like Intel is some small company without access to the same process eqpt, without funds to pay for whatever it takes, without access to whatever it takes. They were ahead for decade s, so why is only TSMC capable now? I'm sure books will be written about this in the years to come. My best guess is that it's the result of bad managem ent, choosing accountants to run a leading edge, high tech company.

Except for the fact that Qualcomm doesn't actually manufacture any CPUs at all.

n era in US tech

ago, Intel held out, a strategy that is now in tatters, with the company? ??s factories struggling to keep up."

ings

screwed

m.

on

is

ust

tel.

des,

is

ement,

It's the need for custom clean room. Both Samsung and Intel are under-esti mating the yield issues. TSMC spend lots of time and effort in cleaning, a nd taking Samsung and Intel to the cleaner. I heard that TSMC is 100x clea ner than the competitors.

The 8086 was not even really ugly once you had grogged that the segment registers were no registers but the MMU. It was a one-address machine that was easy to maintain and was just twice the number of transistors of the Z80 (16000 vs 8000). The 68K had 4 times the number of transistors of the 8086, and they had a s***ty yield. It was on the border of impossible. At that time, Intel's slogan was "Intel delivers" an that got them the IBM design win. IBM took what existed _really_.

Moto kept adding features in the style "double memory indirect + scale + bla" that nobody needed and finally it was easy to write an instruction that might never end, with segmentation violations, page faults and other exceptions all in a single instruction and maybe with multiple memory accesses. That was the time when they knew they could no longer support this.

Intel never succumbed to this featurism. They removed the real drawback, and that was the small memory space and concentrated on computation power.

Oh-so-wonderful-and-so-much-better Intel roadkills:

It seems, Intel's X86 decisions were better than average.

Today, X86 is just a multiple RISC with hundreds of renaming registers and a 860 style instruction bundle decoder that smells like X86. You won't find an EAX in it.

And this legion of ARMs are mostly simplistic CPUs. The few tries with speculative execution had their share of problems too.

Gerhard

Americans can always be counted on to do the sensible thing, after all other options are exhausted.

It tends to become more apparent when something doesn't work right and you have to figure out why. And find that it's something with the fuel pump, say. And find that on the early 2000s Malibus, a car you have, they mount the fuel pump on top of the fuel tank and you have to remove the whole fuel tank to get at it. Now, to remove the fuel tank you have to remove the....

Unnnnnuggggghh

It ran Windows

That it had too few general-purpose registers to make writing efficient compilers more straightforward was pretty much fixed in x86 64, yeah, 30 years later.

I don't see ARM on that list... I'm just sayin'...

Yeah, they are "simple" in comparison to what the x86 has become, but they do the job required and in a much more power efficient way. The point is the desk/laptop arena is becoming a niche while the mobile market continues to grow.

It is highly concerning that Intel has lost the cutting edge in process technology. That's a huge indicator of a fundamental change in the market of CPUs and possibly digital IC design in general.

If companies like Intel can't maintain a fundamental advantage by building their own processing facilities how much longer will the contract manufacturers be able to continue to advance the art?

Or PowerPC.

It killed most of the above also in the Unix arena. When you wanted fast Modelsim in large companies maybe

ISIS is Islamic State of Iraq and Syria, there is no Iran part in it.

Best regards, Piotr

Am 29.07.20 um 19:58 schrieb bitrex:

It wasn't much worse than the PDP11 that also had only 8 registers, including PC, Stack pointer, and one for the base pointer.

For a compiler, you need 0, 1 or infinite resources. Anything else is just an allocation problem.

And that game with the renaming registers began with the 200 MHz Pentium Pro that shocked the Alpha, PowerPC world and the others who considered themselves leading.

The registers do not need a name, it's enough that they are there and that the hardware keeps track of what is where.

of an era in US tech

ears ago, Intel held out, a strategy that is now in tatters, with the compa ny?s factories struggling to keep up."

earnings

irst screwed

nwhile

it has

Intel

g 7nm

n 3 nm.

d.

duction

this is

ies.

t's just

would

likely

right

ch Intel.

em

es

out

new

ARM

ad

e,

sly

gant. Wow. And IDK how you make something inherently elegant, whatever th at even

internally a RISC design and emulate the x86 instruction set via microcode.

CISC processors have used microcode all along. And it would be incredibly stupid to go backwards, to use microcode to execute instructions in multiple steps, to emulate something, instead of executing them directly.

BTW, you're the genius that was arguing with me a couple years ago that having an accountant from Ebay as Intel's CEO was a swell idea, because it's just all about management. Doesn't look so smart today, does it?