Intel seems to be bleeding out ...

[brucehoult]

I've watched Jim Keller talk a few times, and I can never figure out if his key skills are technical, organisational or BS spotting. He certainly seems to have had a very positive impact wherever he has gone.

Yeah. He also had a hand in VAX 8800, Alpha 21164 and 21264, AMD Athlon, the design of the amd64/x86_64 ISA, co-founder of PA Semi (developing a PowerPC CPU) followed by a stint designing Arm cores for Apple when they bought PA Semi. And Tesla.

Like Musk, Keller has a knack of separating the things that violate the laws of physics from the things that don't, and then saying "Ok, that *should* be possible ... now how do we achieve it?"

[mengfei]

..and cutting loss continues

https://www.thestar.com.my/business/business-news/2024/09/04/intels-new-plant-on-partial-pause

https://www.nasdaq.com/articles/intel-intc-eyes-strategic-overhaul-potential-sell-altera

Any takers?

[Marco]

It's a shame that Intel has all this trouble right as their processors and GPUs are doing a lot better. Then again, that just shows how bad their process is doing ... buying the best TSMC process and system design are important, ISA is almost irrelevant with laptop or even tablet formats.

https://www.pcgamer.com/hardware/gaming-laptops/these-new-asus-lunar-lake-laptops-with-27-hours-of-battery-life-kinda-prove-its-not-just-x86-vs-arm-when-it-comes-to-power-efficiency/

[David Hess]

... ISA is almost irrelevant with laptop or even tablet formats.

Do you mean as far as power?  ISA sure matters for applications.

[magic]

Well, an increasing percentage of users download those from "app stores" and don't care
(On tablets it's pretty much 100%).

And there is emulation. Sure, it's not as fast as native, sure it won't run everything. But I have a Linux ARM laptop and it runs LTspice fast enough to be usable, or Wintel applications for industrial stuff control over RS232 using USB dongles. I haven't booted actual Windows in over a year.

[David Hess]

I still use some of the same applications that I used 20 years ago; AutoCAD 2000, Forte Agent, Mathematica, etc.  If the computer does not run my old applications, then I am not interested.

[Doctorandus_P]

I have not bought an Intel processor ever since they forced shops to not sell AMD processors. But I did not keep up with the news, I'm not even sure whether they ever got fined for that.

[David Hess]

I have not bought an Intel processor ever since they forced shops to not sell AMD processors. But I did not keep up with the news, I'm not even sure whether they ever got fined for that.

I stopped when they dropped all support of ECC on consumer CPUs.  My last Intel system was a Pentium4 2.4C with 875P chipset.

In theory some Intel consumer stuff supports ECC now, but it is a maze to figure it out with official sources giving conflicting information.

[Doctorandus_P]

That pentium 4 thing may have been around the same time.

Also triggers a memory of the intel compiler, at some time, compiled programs ran about 30% or so slower if it could not find the string "genuine intel" somewhere in the registers. It took quite some time before the benchmarkers figured out what was happening. Heck, just giving their competitors newest CPU some undeserved low benchmark results in the first month or two after release must be worth millions or billions.

... And that coprocessor calculation bug that gave wrong answers... On itself it should have been just an errata, but they tried to cover it up untill such an outrage was started in the media that they could not deny it any longer.

[David Hess]

You can read the story here:

https://en.wikipedia.org/wiki/Intel_C%2B%2B_Compiler#Support_for_non-Intel_processors

Note that toward the end, the link says:

The FTC settlement included a disclosure provision where Intel must:[18]

    publish clearly that its compiler discriminates against non-Intel processors (such as AMD's designs), not fully utilizing their features and producing inferior code.

In compliance with this ruling, Intel added disclaimers to its compiler documentation:[19]

Intel posted the disclaimer as a graphics image so that search engines could not find and index it.

[radiolistener]

Just read that they will be working on government contracts. It's a pity, I liked Intel processors. In a way, it's symbolic that Intel is falling apart after the discontinuation of the Z80.

[iMo]

Just read that they will be working on government contracts. It's a pity, I liked Intel processors. In a way, it's symbolic that Intel is falling apart after the discontinuation of the Z80.

Intel is not falling apart.. It just undergo a facelift. They will get rid of the FPGA business, they will sell the production sites to somebody (selling it to TSMC would be a bad move, imho, as TSMC's market share is already too large to be safe considering the geopolitics, new manufacturers in US and Europe are required), and they will cut the workforce to something like 50k within next 5 years. That gives Intel new fresh wind in their sails for working on their architecture improvements..

[radiolistener]

Government subsidies have never provided wind in the sails, on the contrary... And now it's not a surprise that they rolled out defective CPU...

[Marco]

selling it to TSMC would be a bad move, imho, as TSMC's market share is already too large to be safe considering the geopolitics, new manufacturers in US and Europe are required

TSMC gets so much better revenue out of every EUV machine, who is supposed to buy it?

Intel and Samsung can to a certain extent subsidize their fabs, but a pure foundry would have to compete on even playing field with TSMC ... and if Intel and Samsung can't do it, what is anyone else's chance?

[coppice]

Intel is not falling apart.. It just undergo a facelift. They will get rid of the FPGA business, they will sell the production sites to somebody (selling it to TSMC would be a bad move, imho, as TSMC's market share is already too large to be safe considering the geopolitics, new manufacturers in US and Europe are required), and they will cut the workforce to something like 50k within next 5 years. That gives Intel new fresh wind in their sails for working on their architecture improvements..

What possible value could Intel's fab network have for TSMC? It wouldn't buy them a bunch of new customers, and those fabs would be a dead weight around their neck.

[TimFox]

Just read that they will be working on government contracts. It's a pity, I liked Intel processors. In a way, it's symbolic that Intel is falling apart after the discontinuation of the Z80.

The Z80 was developed by Zilog, originally a subsidiary of Exxon (the oil company) and was designed to be an improved version of Intel's 8080 series.
Zilog discontinued the Z80 in April of 2024;  however, they still make the eZ80.
Zilog is now part of Littelfuse, along with IXYS.  https://www.zilog.com/

[iMo]

Intel is not falling apart.. It just undergo a facelift. They will get rid of the FPGA business, they will sell the production sites to somebody (selling it to TSMC would be a bad move, imho, as TSMC's market share is already too large to be safe considering the geopolitics, new manufacturers in US and Europe are required), and they will cut the workforce to something like 50k within next 5 years. That gives Intel new fresh wind in their sails for working on their architecture improvements..

What possible value could Intel's fab network have for TSMC? It wouldn't buy them a bunch of new customers, and those fabs would be a dead weight around their neck.

TSMC is investing into 12-28nm nodes these days, afaik (ie the ESMC in Dresden, 10billion Euro, 70% owns TSMC), so why not to buy Intel's fabs?

[radiolistener]

The Z80 was developed by Zilog, originally a subsidiary of Exxon (the oil company) and was designed to be an improved version of Intel's 8080 series.

It was based on the Intel 8080 architecture. It was a well-known architecture and wonderful success at the time. The Zilog Z80 gained its popularity due to its software compatibility with the Intel 8080. Now, the Z80 is discontinued, and Intel has also falling apart. The era of Intel and success of their architecture is coming to an end. This is what I mean...

[David Hess]

The Zilog Z80 gained its popularity due to its software compatibility with the Intel 8080.

Software compatibility and an enhanced instruction set had little to do with the Z80's popularity.  What the Z80, and 8085, did have was easier interfacing and a requirement for only a single +5 volt supply.  The 8080 required multiple supply voltages and a bunch of external logic for interfacing making it more difficult and costly to use.

We switched to using the 8085 for embedded use as soon as we could to simplify hardware.  All of our development systems used the Z80, but without any Z80 specific applications.

[radiolistener]

What the Z80, and 8085, did have was easier interfacing and a requirement for only a single +5 volt supply.

Yes, no need for negative voltage was a significant advantage of the Z80. However, I believe this was more due to the limitations of silicon refinement processes at the time the 8080 was developed. If the 8080 had been designed later, it likely would have also operated with a single 5V supply.

[coppice]

What the Z80, and 8085, did have was easier interfacing and a requirement for only a single +5 volt supply.

Yes, no need for negative voltage was a significant advantage of the Z80. However, I believe this was more due to the limitations of silicon refinement processes at the time the 8080 was developed. If the 8080 had been designed later, it likely would have also operated with a single 5V supply.

For many people the biggest benefit of single rail MOS was not avoiding the need for the 12V and -5V rails, but avoiding the need to have them sequenced, so the -5V substrate bias was present first.

[PA0PBZ]

https://www.cnbc.com/2024/09/20/qualcomm-reportedly-approached-intel-about-takeover.html

 

[SiliconWizard]

Yes, this is getting fun.

[coppice]

TSMC is investing into 12-28nm nodes these days, afaik (ie the ESMC in Dresden, 10billion Euro, 70% owns TSMC), so why not to buy Intel's fabs?

Its a popular strategy to buy fabs which are somewhat away from the bleeding edge, and make good money from the low investment it takes to buy and run a fully depreciated fab. Most of the leading edge fab people have sold off fabs in this way, typically to people in the analogue or power areas, who have no need for the finest geometries. Intel has tended to follow a path of leading edge fabs for CPUs, ageing them into fabs for chip sets, ageing into the least geometry sensitive stuff they do, and then selling things on to other vendors who don't need fine geometries. The key thing with all these transfers is the receiving party gets a real bargain. I don't think Intel is looking to raise a little money with bargain basement selloffs of older fabs. If you look at what TSMC is doing for 28nm, you may already find a lot of second hand equipment among their purchases. It really depends on the market of the moment, and which types of IC are hot right now. If 28nm is hot enough the only available equipment may be new.

[David Hess]

Besides analog, digital logic which relies on floating gate memory like EEPROM and Flash is limited to larger feature sizes.  This applies to a lot of microcontrollers and some programmable logic.  I think NOR Flash is just recently down to 28 nanometers?