The Fall of Intel

[Cmaier]

Interesting: https://www.nytimes.com/2024/09/20/technology/qualcomm-intel-talks-sale.html

We have another thread on this:

Thread 'qualcomm wants to buy Intel'

Sep 20, 2024

Qualcomm wants to buy Intel

What’s next for Intel?

www.theverge.com

The fanfic on this one is spicy.

• Cmaier
• Replies: 11
• Forum: Technology

 

[dada_dave]

Jason Lefkowitz (@jalefkowit@vmst.io)

In retrospect, it's kind of crazy that the two biggest developments in CPU technology over the last 20 years were "use much less power, for mobile" and "use much more power, for AI", and Intel managed to miss both of them

vmst.io

 

[mr_roboto]

@dada_dave One of the replies to that toot...

Marty Fouts (@MartyFouts@mastodon.online)

@jalefkowit@vmst.io Back in the early days of mobile I was a principal engineer at a well known mobile device company. Intel shared their power targets for their coming “low power” SOC. Their target for idle consumption was higher than the maximum consumption of our existing ARM based product...

mastodon.online

 

[dada_dave]

Intel’s entire turnaround plan hinges on this one new chip family – Clearwater Forest pictured, Intel’s first 18A chip slated for high-volume manufacturing

Clearwater Forest is key to Intel’s future.

www.tomshardware.com

Forget Lunar Lake; If there's one chip family that serves as the lynchpin for Intel’s entire turnaround plan, it's this – the company’s coming Clearwater Forest Xeon that it revealed for the first time at a recent event. The chip family is exceedingly important because this is the first high-volume chip to be fabbed on the Intel 18A process node, a node so critical that Intel CEO Pat Gelsinger has said he has bet the entire company on it.

If Intel can produce those chips economically, they will be the key to its turnaround, particularly because it will entice other companies to have their own chips made at Intel Foundry. If Intel can’t punch these chips out in high volume or is beset by delays, we could be looking at a future that doesn’t include an Intel that resembles its current form.

 

[Cmaier]

That second quote makes no sense to me. Nobody says to themselves “intel was able to produce its own chips economically, so i am enticed to make my chips on Intel’s foundry.”

Foundry decisions are much more complicated than that, involving whether the fab can presently achieve performance targets, price of doing so (a function of yield and other things), and the features and capabilities of the process (e.g. “i need power and ground planes for my design - is there a maximum area design rule that would prevent that?”). It also, often even more importantly, takes into account the future: what will happen to prices over time, what will the next node look like, will they be able to launch the next node on time and meeting my performance/cost requirements, or am i going to have to switch back to TSMC after this round?

Important to understand: it’s going to take a sustained period of good execution before very many people trust their designs to Intel’s fabs. This is particularly the case because Intel doesn’t speak the same language as the rest of the industry, doesn’t use the same EDA tools as the rest of the industry, and is not historically setup to providing design teams outside of Intel with what they need.

 

[dada_dave]

That second quote makes no sense to me. Nobody says to themselves “intel was able to produce its own chips economically, so i am enticed to make my chips on Intel’s foundry.”

Foundry decisions are much more complicated than that, involving whether the fab can presently achieve performance targets, price of doing so (a function of yield and other things), and the features and capabilities of the process (e.g. “i need power and ground planes for my design - is there a maximum area design rule that would prevent that?”). It also, often even more importantly, takes into account the future: what will happen to prices over time, what will the next node look like, will they be able to launch the next node on time and meeting my performance/cost requirements, or am i going to have to switch back to TSMC after this round?

Important to understand: it’s going to take a sustained period of good execution before very many people trust their designs to Intel’s fabs. This is particularly the case because Intel doesn’t speak the same language as the rest of the industry, doesn’t use the same EDA tools as the rest of the industry, and is not historically setup to providing design teams outside of Intel with what they need.

I very much agree with a slight quibble that I think Intel did switch to the same EDA tools as the rest of the industry.

 

[Cmaier]

I very much agree with a slight quibble that I think Intel did switch to the same EDA tools as the rest of the industry.

I don’t think that’s true, at least not across-the-board. While I believe they support those tools (Cadence, synopsys, etc.) for foundry customers, from what I’ve heard they still use their own mess of tools when designing their own chips for their own fabs.

 

[dada_dave]

I don’t think that’s true, at least not across-the-board. While I believe they support those tools (Cadence, synopsys, etc.) for foundry customers, from what I’ve heard they still use their own mess of tools when designing their own chips for their own fabs.

Really? Huh I thought they had switched completely. One would’ve thought going to TSMC N3B for Lunar Lake/Arrow Lake would’ve necessitated such a switch.

But regardless, I agree that 18A being a success for Intel does not imply it’ll be immediately successful as a foundry node. It could take Intel awhile to even get volume available for customers just for starts. Then everything else you said.

 

[dada_dave]

Intel is on track to receive its $8.5 billion CHIPS Act award by year-end

This is a welcome change of pace for Intel.

www.tomshardware.com

I recognize the geopolitical reasons we need to do this, but it is in some ways subsidizing a currently failing business model, probably (and reportedly) the main reason why the government is hesitant to pull the trigger on releasing the funds.

 

[dada_dave]

Samsung has no plans to spin off its critical foundry business

Remains committed to grow foundry business.

www.tomshardware.com

This is about Samsung but the parallels are undeniable. The one thing that confuses me is the following point:

In 2017, Samsung separated its chip manufacturing operations from its design unit to address customer concerns about potential conflicts of interest. However, some clients still worry that the foundry and design divisions may share sensitive information, which undermines confidence in Samsung's neutrality.

No one seems particularly concerned or at least voicing those concerns with Intel as a foundry. Is it because Intel is x86 and most potential customers will be ARM? But most of chip design isn’t really dependent on that and the foundry will make more than just CPUs. Then again Intel did make foundry its own unit as apparently Samsung has done. The only other thing I can think of is that the Samsung design and foundry units used to be combined so this is holdover concern from that but Intel famously had right links between foundry and design. Unless Samsung has actually been caught doing this? Sharing info?

 

[Cmaier]

No one seems particularly concerned or at least voicing those concerns with Intel as a foundry.

Really? That’s not what I hear. It’s not like people are going to publicly announce stuff, but the fact that Intel’s foundry business isn’t getting a lot of takers is partly because Intel is historically untrustworthy.

 

[dada_dave]

Really? That’s not what I hear. It’s not like people are going to publicly announce stuff, but the fact that Intel’s foundry business isn’t getting a lot of takers is partly because Intel is historically untrustworthy.

I haven’t seen it in print the way I just saw it written about for Samsung. Or at least I don’t remember seeing it. And yet if it’s a problem for Samsung then yes it should be a concern for Intel as well I would think. So I don’t doubt your statement.

 

[dada_dave]

TSMC is not interested in buying Intel's fabs

Why buy fabs if you have your own?

www.tomshardware.com

Intel's struggles to land customers for its Intel Foundry unit raised concerns over viability of Pat Gelsinger's IDM 2.0 strategy and resurrected rumors about the company's possible plans to spin off or sell its manufacturing capacity. But only a few companies worldwide can potentially afford and operate Intel's fabs. One of them is TSMC, and it says it is not interested in acquiring those facilities.

 

[Cmaier]

TSMC is not interested in buying Intel's fabs

Why buy fabs if you have your own?

www.tomshardware.com

US would never allow such a transaction anyway. Global Foundries is the most likely possible purchaser.

 

[dada_dave]

US would never allow such a transaction anyway. Global Foundries is the most likely possible purchaser.

Do they even have the money to buy/operate Intel’s foundries? Also GloFo rather pointedly got out of the leading edge race, would they want back in?

 

[Cmaier]

Do they even have the money to buy/operate Intel’s foundries? Also GloFo rather pointedly got out of the leading edge race, would they want back in?

oh, i have no idea if they’d have any interest or financial ability. Just saying they’re the only one that would get approval.

The reason global foundries got out is that they couldn’t compete. I think they’d love to get back into higher margin stuff.

I suppose, in theory, you could imagine other U.S. companies that could theoretically have interest - Broadcom, or Micron, or whatever. They are obviously worse fits. Maybe a consortium - Apple, AMD, Oracle, Broadcom, nVidia, etc. Would never happen, of course, but fanfic is always fun.

 

[Yoused]

I read that 18Å had some delightful features involving elaborate layering and backside power delivery, but Lunar Lake seems to be on N3, so Intel must still be struggling to get the process reasonably stable.

 

[dada_dave]

I read that 18Å had some delightful features involving elaborate layering and backside power delivery, but Lunar Lake seems to be on N3, so Intel must still be struggling to get the process reasonably stable.

I don’t think 18A was ever going to be this early but there were supposedly plans for its sister node 20A to produce some Arrow Lake dies though those plans were apparently shelved and now everything this generation is TSMC 3nm.

 

[dada_dave]

Some rare good news for Intel:

Intel stock on the rise as EU anti-trust probe ends in nothing

After escaping a fine of over $1.1 billion, Intel is now free of all the legal trouble connected to the nearly two-decade-long fight with EU regulators, thanks to a decision of the EU Court of Justice. The tech giant had been accused of trying to block AMD by awarding rebates to Lenovo, Dell...

www.notebookcheck.net

Also shows that European Courts can disagree with the European Commission on the matters of application of the law, something to consider if/when the EU commission makes regulatory judgements against Apple. They will not necessarily be upheld. That said, in this particular case I am a little surprised as I'm pretty sure the US successfully fined Intel for this exact practice and its not exactly like the US is known for stronger anti-trust measures in general. Also of note, the final judgement in this case today stems from a fine from 2009 and a General Court decision in 2022. Holy shit. And I thought US court cases wormed their way through the system slowly.

 

[Cmaier]

Some rare good news for Intel:

Intel stock on the rise as EU anti-trust probe ends in nothing

After escaping a fine of over $1.1 billion, Intel is now free of all the legal trouble connected to the nearly two-decade-long fight with EU regulators, thanks to a decision of the EU Court of Justice. The tech giant had been accused of trying to block AMD by awarding rebates to Lenovo, Dell...

www.notebookcheck.net

Also shows that European Courts can disagree with the European Commission on the matters of application of the law, something to consider if/when the EU commission makes regulatory judgements against Apple. They will not necessarily be upheld. That said, in this particular case I am a little surprised as I'm pretty sure the US successfully fined Intel for this exact practice and its not exactly like the US is known for stronger anti-trust measures in general. Also of note, the final judgement in this case today stems from a fine from 2009 and a General Court decision in 2022. Holy shit. And I thought US court cases wormed their way through the system slowly.

This took 15 years. That’s a very long period of uncertainty for every company involved. The EU is a horrible place to do business. Regulatory uncertainty is not a good thing.