What went wrong?

Also, the prediction that x86 CISC wasn't going to be able to increase in performance was wrong.
If x86 actually had run out of performance, things might have turned out differently.
The closest we got to that prediction was the Netburst architecture.

Attached: 1559575918879.png (610x1054, 160K)

AMD made 64-bit mainstream, which made powerful computers accessible to regular users.

Look, one day they are gonna have to pick a team. You can't flip sides forever, one day you're gonna have to settle down and you'd better make up your mind before that day comes

They would use old fabs.

x86 really IS a hell of a lot harder to speed up. But when everyone else makes a series of fatally stupid business decisions, the inherent architectural differences don't get much chance to show.
It turned out that "Intel is spending billions more on fabs than us, so let's abandon the desktop/laptop market and focus on servers" was the wrong decision for most companies.

You're right.
The question now is how long is the x86 train going to go?
It's a Frankenstein of an architecture and I'd like to see something new, and well architected, replace it
but it will be worse because NOBODY will let a 2nd source implement the architecture (without onerous licensing like ARM does)

>inb4 someone says RISC-V

>x86 really IS a hell of a lot harder to speed up.
Are they really, though? For a long time x86 from both intel and AMD are RISC cores running CISC code through a decoder. Back when PPC started-up x86 was pure CISC, but nowadays they're improving the RISC cores and translation efficency.

>so consoles switch to x86/ARM 8th gen
they switched to allow crapware on their stores
to be just like steam

Apple sabotaged the G5 so they could make the switch to intlel

>Are they really, though?
We are.
The performance jump from 8088 to 286 was huge
Similarly, 286 to 386 was also huge
The jump to 486 wasn't as big, but caching made it better
The jump to the pentium added superscaler architecture and was the last really significant performance jump until the core series of chips replaced the netburst chips
Hyperthreading barely had any performance impact
the jump to 64bit gave 0 performance benefit

Not until multi core did any significant performance improvements happen. They were all just iterative improvements.

Now we are in the core wars. 3rd gen ryzen gets most of its performance increase from cache, and Intel can't move below 14nm. CPUs are drawing over 100w and all we can do is add cores, as its looking like 5GHz is also a major obstacle.

The performance we got from optimizations of RISC to CISC translations is coming to an end. The diminishing returns keep piling up, and things like SPECTRE/MELTDOWN/MDS/ZOMBIELOAD etc are even undoing some of those performance improvements.
Scaling up with cores is all we got. But at some point we're going to have to architect.

>Wrong
It went just right, PowerPC deserved that since 68k was killed to make room for it.

>Are they really, though? For a long time x86 from both intel and AMD are RISC cores running CISC code through a decoder.
Yes. It's still bad because a lot of that baggage still propagates all the way through the internally-RISC-like levels. Some of it is inherent bottlenecks of the instruction set and variable-width instructions, some of it is layers of extra mandatory buffer/table bloat and additional scheduler complexity. That also makes it harder to extract more parallelism from the instructions, and it makes it harder to add more threads because every additional thread needs to duplicate that bloated decoder and tie into the extra-complicated scheduler. Even when they downplay the decoder penalty as much as possible, they still admit it's at least a few percent of the transistor count. IIRC it used to be like 15%, and the percent only crept down over time because L1/2/3 caches grew faster than the other stuff grew.

One big result of this complexity penalty is that x86 still struggles with hyperthreading, whereas a bunch of other RISC architectures have had 4-8 threads per core for years now (POWER9 and Sparc being the examples that come to mind). If the Alpha hadn't been cancelled, it would have actually had 4 threads back in like 2004. 4 roughly-equal threads, too, not like x86's 15-40% performance on the second thread.

The other big thing that comes to mind is how the Atom utterly failed to compete with ARM; because even with all the high end optimizations stripped out, a minimal implementation of x86 is still like twice the size of a comparable ARM implementation. The Atom sucked down like twice the power even when Intel had a full generation fab lead.

I have a good feeling about POWER10.
It might be their comeback.

That's like saying its not even x86 anymore its just x64

x64 is still LE

That’s complete and utter nonsense. RROD was ATI’s fuckup.

Anyone asked for cracked solder joints and tin whiskers? we got both in spades here!

But we removed the lead, so the massive pile of dead electronic devices we caused is harmless (tm)

Attached: rohs.gif (630x630, 53K)

I guarantee RoHS has caused more deaths than lives it has saved
There is not a single reason to get rid of lead on solder other than acceleration of obsolesce
Hardware deemed "critical" is allowed to use leaded solder, yet there is no conclusion that leaded solder is more reliable so someone is not telling the truth

It's probably one of the dumbest shits we ever did.
Not Thalidomide bad, but quite close.

Unless they target consumers are reasonable prices, which IBM won't, they will continue to be a corporate company only.

Apple dropping PowerPC was such a disappoint. POWER now has nearly 200 threads on its top processor, they supported PCI4 before anyone else, they support openCAPI which is even better than PCI4. The I/O it can do is insane. God damnit.

What is the likelihood of apple dropping x86 to return to Power? That would be interesting. They’d lock it down like fucking crazy though.

So many devices in the early to mid 2000’s suffered at the hands of lead free solder. iBooks, game consoles, thinkpads, etc..

The engineering firm they bought to work on their ARM designs also had a Power core. en.wikipedia.org/wiki/P.A._Semi
I don't think they'll switch back though.

If they do Switch it will be to their own A chips.

Not only did they have their own G5 implementation, but it was also the solution to the G5 laptop. Sadly, Apple was already too far into the Intel transition for that to save PowerPC

IBM had a mobile chip in te works too.: The 970GX

Nigger the military still uses pentium 2 and even Z80 insome of its equipment. It's not because shit is good, but because of its already extensive documentation

>The 970GX
It was too slow.
IBM promised Jobs a 3GHz PPC chip. He told the board at Apple and then the faithful at the next WWDC. Then they stiffed him. Too busy with games machine makers to be bothered with a couple of million special PPC chips.
The Wrath of Jobs was something to see. He was on the phone to Paul Otellini and the deed was done.

>3GHz G5
That was a separate issue, nobody was expecting a 3GHz G5 in a laptop.

>That was a separate issue, nobody was expecting a 3GHz G5 in a laptop.
But the effect was the same.
Once Jobs was mad at IBM he ditched PPC.

Thank you for good informative posts. It's pretty rare to see things like that on Jow Forums, in nowadays.

I wonder if x86 could have saved Amiga

>I wonder if x86 could have saved Amiga
Amiga needed to come out before the Macintrash in order to have had a foothold.
Also, they should have adopted PC style (cheap) cases, peripherals, etc.
That and Commodore kept blowing money on failed systems like the C16.

>It turned out that "Intel is spending billions more on fabs than us, so let's abandon the desktop/laptop market and focus on servers" was the wrong decision for most companies
Yeah, focus on servers then sink their entire server lineup to the Itanic.

>Once Jobs was mad at IBM he ditched PPC
It was IBM who was mad at Applel because of their manufacturer warehouse rot practices