The Fall of Intel

[Artemis]

This is why I don’t take this stuff seriously. The reason Wall Street ousted Gelsinger is much more likely to be very simple: they’re impatient and there is simply no world where Intel is going to turn a ton of profit anytime soon short of shuttering or slowing fab capex and milking the design unit on TSMC wafers (and even older nodes) with incremental improvement to slow the advance of Arm laptops & AMD in servers.

I’m sure he wasn’t perfect but people are reading too much into this. Turning around Intel is more like turning around Boeing than it is idk, what Ballmer and Nadella have done for Microsoft on the enterprise end or whatever. At least if you actually care about the fabs. Otherwise they can milk it like IBM until they’re acquired for nothing in the 30’s, which to be honest will happen to the design unit regardless, it’s just a question of where their fab orders and revenue go in the short term. I’d like to see that burnt off towards something actually good rather than the shareholders here.

 

[dada_dave]

I’m skeptical of this. Panther Lake is going to be on 18A and while die dize is related to yield and node differences there are less obvious than with larger dice, this is missing something. Panther Lake is less tiled than Meteor Lake — it should be a CPU + SoC tile on 18A and GPU on N3 — and that 18A tile won’t be too far off from the dice used in a phone either + or -.

Of course it could be another 10nm I just wouldn’t take this too seriously. No clients that have *officially* announced their use of 18A rather than interest have backed out. What we do know is the HD cells aren’t good enough yet on density and leakage, which is why 18AP is coming to amend that with Intel explicitly noting it for mobile, and this is also the reason Qualcomm apparently dropped early interest.

View attachment 33030
Simplified $20,000 wafer cost adjusted for various dice. In practice it might be better than this if your marginal area is a lot of parallel logic with redundancy (e.g. SoC blowing up from 150mm^2 to 250mm^2 where the main difference is GPU and encoding size) but still.

It’s also a South Korean outlet. Keep in mind Korea’s AI firms are ditching Samsung and Exynos is a disaster while at least Intel can claim they use their process and Microsoft is lined up for an ASIC or CPU or whatever it was on 18A.

Very possible. The one thing that gives it credence is that Reuters report linked in the article that at least one customer, Broadcom, was unhappy with Intel’s 18A progress and IIRC poor yields were thought to be the issue. So there is supporting evidence. Further, this time last year, Arrow Lake was still reportedly using 20A which obviously didn’t happen. As aforementioned though, even if yields are bad now, and they may not be, the node isn’t supposed to be available for mass production yet anyway so yields have time to improve.

My main concern is that if the report is accurate then that Samsung also struggled with GAA yields could indicate an intrinsic issue with GAA fabrication for mass production that TSMC will also hit even though they’re being more cautious in their timeline to bring it to market. Of course the report could be wrong and/or TSMC’s more cautious approach may pay off where their GAA N2 node is fine.

This is why I don’t take this stuff seriously. The reason Wall Street ousted Gelsinger is much more likely to be very simple: they’re impatient and there is simply no world where Intel is going to turn a ton of profit anytime soon short of shuttering or slowing fab capex and milking the design unit on TSMC wafers (and even older nodes) with incremental improvement to slow the advance of Arm laptops & AMD in servers.

I’m sure he wasn’t perfect but people are reading too much into this. Turning around Intel is more like turning around Boeing than it is idk, what Ballmer and Nadella have done for Microsoft on the enterprise end or whatever. At least if you actually care about the fabs. Otherwise they can milk it like IBM until they’re acquired for nothing in the 30’s, which to be honest will happen to the design unit regardless, it’s just a question of where their fab orders and revenue go in the short term. I’d like to see that burnt off towards something actually good rather than the shareholders here.

I disagree. If Gelsinger had been under pressure from "activist investors" and routinely suffering votes of no confidence in shareholder meetings then I would agree that this was Wall Street mucking things up for short term profit. But this came out of the blue (well at least to us, we don't know how contentious board meetings were before this, see the edit below) after a meeting with the board after the board and the company had expressed massive confidence in Gelsinger by abjuring their retirement rules to keep him long term. All that to me indicates something substantial in Gelsinger's plan to turn Intel around failed badly enough that they had to take this step. Maybe it was just the AI progress or lack thereof, but it could be the fabs.

EDIT:

This is potentially interesting:

Intel's next CEO could be Lip-Bu Tan, a former member of the board who reportedly clashed with Gelsinger about strategy

No decision has been made yet.

www.tomshardware.com

Although no decision has been made, Lip-Bu Tan, a veteran investor in the technology industry and former CEO of Cadence Design Systems, has emerged as a potential candidate. Tan previously played a key role in transforming Cadence into a thriving business and has a lot of respect in the industry (he even has imec's Lifetime of Innovation Award), which makes him a strong contender.

Lip-Bu Tan served on Intel's board for nearly two years after joining it in September 2022 and later taking on expanded responsibilities related to manufacturing operations. During his time on Intel's board, Lip-Bu Tan reportedly clashed with Pat Gelsinger over various issues, including workforce size, the company's strategy for contract manufacturing, and its internal culture. These disagreements eventually led to Tan's departure in August 2023[2024]. Despite this, Intel's board has recently reached out to Tan to explore his interest in leading the company, Reuters reports.

EDIT2:

Also this:

TSMC's N2 process has a major advantage over Intel's 18A: SRAM density

SRAM is hard to scale.

www.tomshardware.com

SRAM is just one metric of course. Hopefully we'll get comparable designs like the same ARM core on both nodes. That'll be fun for people to deep dive on.

 

[Artemis]

possible. The one thing that gives it credence is that Reuters report linked in the article that at least one customer, Broadcom, was unhappy with Intel’s 18A progress and IIRC poor yields were thought to be the issue. So there is supporting evidence. Further, this time last year, Arrow Lake was still reportedly using 20A which obviously didn’t happen. As aforementioned though, even if yields are bad now, and they may not be, the node isn’t supposed to be available for mass production yet anyway so yields have time to improve.

Possibly fair, I’d have to check. I was under the impression this leak was the exact same as Qualcomm’s issue, they wanted better HD libraries.
Either way, what we literally do know is:

A) Intel realizes this [and mobile or mass embedded or Arm server targeting] is a problem and has a separate process modification for 18A along with tacitly acknowledging this flaw in their own presentations.

And B) Samsung’s nodes are so bad that even Google is passing up on 3NM GAAFET for TSMC N3E next year, and Samsung swapped the 2400’s planned 2/3NM for an updated 4NM LPP+ process (which still only shipped in the 24 Ultra for most), and the 2500 2/3NM 2025 rumors are intact also dicey again, with major reorganizations and crisis inside Samsung’s logic foundry part n/n in the last 5 years recently.

I am pretty comfortable saying Intel has a good shot by the 2026-2028 timeframe for a process that’s good enough for ultra mobile at some discount, which, mind you, we really just don’t have right now. It’s not even about being “as good as TSMC” through and through: you can’t even buy a true TSMC N4P competitor +-10-15 % power/density right now at scale.

The hope with Intel is they at least have some basics down on power/performance and Samsung’s problems predate GAA.

My main concern is that if the report is accurate then that Samsung also struggled with GAA yields could indicate an intrinsic issue with GAA fabrication for mass production

Nah. More than FinFET sure, but FinFET had teething pains too, but it was still amazing. You don’t get this far past research and trials for HVM to be a shot in the dark. I will totally acknowledge if I’m wrong but I’d bet it ends just fine. The main benefit of GAA is an intrinsic advantage to power/performance at the same density/chemical treatment/architecture (just like FinFET offered this over Planar) along with tailoring some geometries in ways not possible with FinFET. And just basic leakage reduction again which is becoming more of power consumption at denser nodes

that TSMC will also hit even though they’re being more cautious in their timeline to bring it to market. Of course the report could be wrong and/or TSMC’s more cautious approach may pay off where their GAA N2 node is fine.

Remember this is Intel and Samsung. And TSMC’s N5 & N3E were really exceptional aberrations, N7/N10 (or 12 whatever) weren’t even as rapid re reductions in defects and stuff even though they were great.

I disagree. If Gelsinger had been under pressure from "activist investors" and routinely suffering votes of no confidence in shareholder meetings then I would agree that this was Wall Street mucking things up for short term profit.

I mean he was under pressure from AI, but yes it didn’t reach that point. Coming out of the blue really doesn’t make much difference to me, the behavior from the street can differ! They have been whining about Intel’s cash flow for a while. Wall Street wanted them to sell the fabs.

But this came out of the blue (well at least to us, we don't know how contentious board meetings were before this, see the edit below) after a meeting with the board after the board and the company had expressed massive confidence in Gelsinger by abjuring their retirement rules to keep him long term. All that to me indicates something substantial in Gelsinger's plan to turn Intel around failed badly enough that they had to take this step. Maybe it was just the AI progress or lack thereof, but it could be the fabs.

I mean it was always dependent on government aid at the end of the day. Gelsinger is a politician and visionary on some level, his ideas here on fabs in the west and an open Intel foundry are living on even if our sloth ends up quashing them.

Turning around Intel for 0-10 year pure financial wins would be selling the fabs off and going to TSMC and Samsung for cheaper stuff indefinitely, and adopting their packaging too. That’s the truth. The fabs were never going to be a short or even medium term financial win. Never. What Gelsinger probably didn’t expect was Intel@/ design to decline as much as it did which means this strategy for the long long run to save the design unit from itself (and Intel’s overall profit) is unsustainable because they cannot fund the capex expansions.

10’s Intel absolutely could have but they were screwing around on bad phone chip contra revenue, manycore GPU imitations, dumb drones, buying Altera, half-assing modems etc

EDIT:

This is potentially interesting:

Intel's next CEO could be Lip-Bu Tan, a former member of the board who reportedly clashed with Gelsinger about strategy

No decision has been made yet.

www.tomshardware.com

EDIT2:

Also this:

TSMC's N2 process has a major advantage over Intel's 18A: SRAM density

SRAM is hard to scale.

www.tomshardware.com

SRAM is just one metric of course. Hopefully we'll get comparable designs like the same ARM core on both nodes. That'll be fun for people to deep dive on.

 

[dada_dave]

Possibly fair, I’d have to check. I was under the impression this leak was the exact same as Qualcomm’s issue, they wanted better HD libraries.
Either way, what we literally do know is:

A) Intel realizes this [and mobile or mass embedded or Arm server targeting] is a problem and has a separate process modification for 18A along with tacitly acknowledging this flaw in their own presentations.

And B) Samsung’s nodes are so bad that even Google is passing up on 3NM GAAFET for TSMC N3E next year, and Samsung swapped the 2400’s planned 2/3NM for an updated 4NM LPP+ process (which still only shipped in the 24 Ultra for most), and the 2500 2/3NM 2025 rumors are intact also dicey again, with major reorganizations and crisis inside Samsung’s logic foundry part n/n in the last 5 years recently.

I am pretty comfortable saying Intel has a good shot by the 2026-2028 timeframe for a process that’s good enough for ultra mobile at some discount, which, mind you, we really just don’t have right now. It’s not even about being “as good as TSMC” through and through: you can’t even buy a true TSMC N4P competitor +-10-15 % power/density right now at scale.

The hope with Intel is they at least have some basics down on power/performance and Samsung’s problems predate GAA.

Nah. More than FinFET sure, but FinFET had teething pains too, but it was still amazing. You don’t get this far past research and trials for HVM to be a shot in the dark. I will totally acknowledge if I’m wrong but I’d bet it ends just fine. The main benefit of GAA is an intrinsic advantage to power/performance at the same density/chemical treatment/architecture (just like FinFET offered this over Planar) along with tailoring some geometries in ways not possible with FinFET. And just basic leakage reduction again which is becoming more of power consumption at denser nodes

True, Samsung has had its problems before but the GAA node in particular has reportedly been disastrously bad. With rumors that Intel is also having yield issues with their GAA node, I'm a little concerned even though I admit that concern is probably an overreaction especially as the Intel rumor is, so far, just a rumor, albeit from multiple sources now.

Remember this is Intel and Samsung. And TSMC’s N5 & N3E were really exceptional aberrations, N7/N10 (or 12 whatever) weren’t even as rapid re reductions in defects and stuff even though they were great.

I mean he was under pressure from AI, but yes it didn’t reach that point. Coming out of the blue really doesn’t make much difference to me, the behavior from the street can differ! They have been whining about Intel’s cash flow for a while. Wall Street wanted them to sell the fabs.

I mean it was always dependent on government aid at the end of the day. Gelsinger is a politician and visionary on some level, his ideas here on fabs in the west and an open Intel foundry are living on even if our sloth ends up quashing them.

Turning around Intel for pure financial wins would be selling the fabs off and going to TSMC and Samsung for cheaper stuff indefinitely, and adopting their packaging too. That’s the truth. The fabs were never going to be a short or even medium term financial win. Never. What Gelsinger probably didn’t expect was Intel@/ design to decline as much as it did which means this strategy for the long long run to save the design unit from itself (and Intel’s overall profit) is unsustainable because they cannot fund the capex expansions.

Except that's the issue though. That the foundries weren't going to turn a quick profit absolutely would've been a known quantity, even to the board and the board went along with Gelsinger's plan ... until they didn't. Keep in mind the board are fellow tech entrepreneurs for the most part, people like the one they're considering for CEO. Yes, they represent the shareholders, but they are not finance bros unaware of how fabs operate. The timing and nature of this firing suggests that something went very wrong. As you say maybe it was just that the decline in revenue from their design unit was larger than expected, which meant even though those divisions still turned a healthy profit when viewed in isolation, the profits from those divisions were not enough to offset the losses from rapid fab expansion even with the government assistance. We'll know more next quarter. While Intel experienced some "one-time" losses last quarter, they may continue to hemorrhage money at an unsustainable rate. Even worse is this rumor that the 18A fabs have yield issues. If that's true ... Gelsinger went with a high-risk high reward strategy that if anything went wrong, he'd be sunk. Given all that, to me, this move is less about Wall Street short term-ism, which I agree can be a problem, and more that it became clear that Gelsinger's turn around plan was no longer plausible and they need a new CEO and new vision.

10’s Intel absolutely could have but they were screwing around on bad phone chip contra revenue, manycore GPU imitations, dumb drones, buying Altera, half-assing modems etc

 

[Artemis]

Like, RE: Samsung worry:

Samsung has been having problems ever since EUV. 4NM*/5NN/7NM were all objectively bad from a parametric yield perspective and Intel too, albeit due to first not adopting EUV and overusing Cobalt in their metal stack for clocks on 10NM (which ruined yields and density + EMI apparently) + trying an insane HD library on 10NM without EUV.

That said, Intel have had problems before though and are still closer to Samsung than TSMC I’d say. TSMC only blew the 20NM Planar node though Apple made it work with the A7.


4NM Samsung is bifurcated: there is a fake 4NM that’s really just 5NM which itself is I think slightly lower density than TSMC and uses much fewer EUV layers, basically like their 7NM +++++ (or some kind of N6 thing) and they used “4NM LPX” as a marketing label for this 5NM process in 2022 with the Snapdragon 8 Gen 1 — then there’s a real “SF4” and 4NM LPP thing which by now at least has Exynos 2400 dice that have ok ish power apparently though the variability is large and it’s clearly still inferior to say the 8 Gen 3 or D9300 with similar cores. The GPU I believe is fine though.

Regardless clear it’s Samsung fabs as the main issue considering Qualcomm would happily take Samsung fabs with -10% logic density, +10-15% power iso-perf or -5% clocks iso-power for a significant discount. And they’d be right to tbh, totally worth it.

What people miss tho is that Qualcomm actually was not happy with how Samsung foundry turned out for the 888 & 8 Gen 1 — this isn’t the tradeoff space anyone looks to make. I bet Nvidia and Qualcomm would gladly do a half or + node trade for significant cost reduction, or trade some variability for cost (which could also be used for more area for SRAM or other power-saving stuff anyway) — but objectively speaking it is accurate to say that is not what the current Samsung offer is one way or another (it could also be volume etc).

Conclusion: Right now I think people both under and over estimate how far ahead TSMC is. On a macro scale for a 10-30 year history of tech, or from a state’s POV, it is overrated by many. From the perspective of a mass market mobile design firm or the full economic POV: still underrated, which is good and bad — only reason I’d say it’s good is it suggests the bar is “low” in a certain sense.

 

[Artemis]

True, Samsung has had its problems before but the GAA node in particular has reportedly been disastrously bad. With rumors that Intel is also having yield issues with their GAA node, I'm a little concerned even though I admit that concern is probably an overreaction especially as the Intel rumor is, so far, just a rumor, albeit from multiple sources now.


Except that's the issue though. That the foundries weren't going to turn a quick profit absolutely would've been a known quantity, even to the board and the board went along with Gelsinger's plan ... until they didn't.

I totally acknowledge that. I’m straightforwardly telling you they were done, and you can’t dial this in to the health of 18A. Even if 18A were a TSMC N3P analogue for mobile it would be met with starter midrange orders before Qualcomm is going to throw a flagship 8 Elite smartphone chip on it. Gelsinger can be seen more of as a long term figure in the way East Asian industrialists would have been in their early days albeit for reform, but that isn’t the system we live in and so there’s only so much federal spigot $.

There might be more now that USG senses a threat to the fabs tbf, but I am literally acknowledging to you I don’t think the entire Intel plan was coherent from a pure private sector perspective on reasonable timescales, 18A being C+ or A+ is irrelevant ultimately. F+ okay, yeah sure that weighs.

Keep in mind the board are fellow tech entrepreneurs for the most part, people like the one they're considering for CEO. Yes, they represent the shareholders, but they are not finance bros unaware of how fabs operate. The timing and nature of this firing suggests that something went very wrong.

As you say maybe it was just that the decline in revenue from their design unit was larger than expected, which meant even though those divisions still turned a healthy profit when viewed in isolation, the profits from those divisions were not enough to offset the losses from rapid fab expansion even with the government assistance.

This would be my guess. The board wanted a turnaround and Intel can BS about the cash even tho fabs are bad for 0-10 year $$$$ milking X86, but it’s too hard to stomach that vs stomaching an ousting of PG when that collapses, and the writing has been on the wall.

The other thing is Intel has failed to get into AI and probably isn’t focusing enough on that.

FWIW, a lot of this (including AI) still predates Gelsinger and every report I’ve seen suggests he was doing a good job of reforming internal culture. I’m not the only one who thinks this was kind of doomed though due to the legacy.

I really believe the only way someone was going to make it through in this:

A) the Wall Street maxi plan: cut the fabs and go as hard as possible on frugal TSMC X86 chips for “milking it” money. They still decline but in silence.
B) Gelsinger, but with much more govt subsidies.

C) Time Machine and undo Intel mistakes in the 10’s.

We'll know more next quarter. While Intel experienced some "one-time" losses last quarter, they may continue to hemorrhage money at an unsustainable rate. Even worse is this rumor that the 18A fabs have yield issues. If that's true ... Gelsinger went with a high-risk high reward strategy that if anything went wrong, he'd be sunk. Given all that, to me, this move is less about Wall Street short term-ism, which I agree can be a problem, and more that it became clear that Gelsinger's turn around plan was no longer plausible and they need a new CEO and new vision.

 

[dada_dave]

I'll admit that this is a little silly, but at risk of trivializing the topic, I've been struck by the parallels between the Intel and, of all things, Manchester United - massive hugely successful entities in their respective fields which seemingly peaked in 2013 but with structural problems already present and then a series of leadership changes none of whom were able to fix the underlying problems (and some who exacerbated them), often burning through CEOs/head coaches leaving multiple visions of what the team/company should be incomplete and thus incoherent - to the point where it almost becomes a "cursed" job. As I said, a little silly, but it's not the worst sports-tech analogy I've ever seen!

 

[Artemis]

The design unit is also still profitable on its own but ofc they are experiencing losses and Lunar Lake is not that economical and they need that style of product, which is funny because it’s not that competitive anyway. It’s going to be an utter sloshing in 2026, 2027, 2028, 2029.

I'll admit that this is a little silly, but at risk of trivializing the topic, I've been struck by the parallels between the Intel and, of all things, Manchester United - massive hugely successful entities in their respective fields which seemingly peaked in 2013 but with structural problems already present and then a series of leadership changes none of whom were able to fix the underlying problems (and some who exacerbated them), often burning through CEOs/head coaches leaving multiple visions of what the team/company should be incomplete and thus incoherent - to the point where it almost becomes a "cursed" job. As I said, a little silly, but it's not the worst sports-tech analogy I've ever seen!

i don’t mind sports analogies at all, have done the opposite with comparing teams to firms and empires etc. imperfect but fun/funny:

 

[Yoused]

I've been struck by the parallels between the Intel and, of all things, Manchester United

Of course, where the analogy starts to break down is that a team which is not doing well in their league is often more profitable than the team with a bunch of fresh cups on display – I think that may have to do with marquee salaries. A company producing a product may do well selling poor-quality stuff, but their success is likely to not persist.

 

[dada_dave]

Of course, where the analogy starts to break down is that a team which is not doing well in their league is often more profitable than the team with a bunch of fresh cups on display – I think that may have to do with marquee salaries. A company producing a product may do well selling poor-quality stuff, but their success is likely to not persist.

Not to side track the thread too much, but not really or at least it depends on what you mean. For a team like Manchester United, contracts lock them into paying those marquee superstar salaries even when those players underperform (though they may not get bonuses depending on their contracts) and missing things like Champions League revenue, or European football at all, or even worse being relegated (which ManUtd is no danger of) can massively impact revenue. While financial fair play rules are incredibly imperfect to say the least, one thing they have accomplished is stopping smaller teams from destroying themselves chasing promotion/Europe which used to happen regularly - because if they failed they couldn't pay their salaries/transfer fees, couldn't offload those players for anywhere near what they had paid for because they had underperformed, and got into a debt cycle which saw them fire sale their team, get relegated, fire sale, get relegated, on and on until eventual liquidation. That very nearly happened to Leeds United who are back in the Premier League and had once been a football power (decades ago) but in the 2000s nearly destroyed themselves chasing Europe and ended nearly exiting the professional football league pyramid altogether after multiple rounds of debt-cycle fueled relegation. Even Barcelona, again never in any danger of being relegated, saddled itself with massive debts, partly because they overpaid for players in transfer and salaries that didn't translate into on the field performance and rewards. Their current team is doing quite well because they managed to hire a good coach who is doing well focusing more on academy players (i.e. young guys who come up through the La Masia academy who are cheaper than buying superstars) which they have to do because of their previously dire financial situation. But the specter of debts and the Spanish football association coming down on them like a ton of bricks is still there.

Now a mid table team that doesn't spend exorbitantly and isn't chasing glory but occasionally gets a boon like European football just by luck can be much more profitable, but there the risk is the opposite, you risk relegation by bad luck as well which can be a massive hit to the club's finances*. Leicester is a great example of this, even won the Premier league unexpectedly (very, very unexpectedly) in 2015 and of course European football, but couldn't sustain themselves. Eventually a bad luck year with multiple players injured and the COVID pandemic, and they suddenly found themselves being relegated a couple of years ago. And only the fact that they had billionaire backers, who ... shall we say creatively got around financial fair play rules, saved them and now they are back in the Premier League (though it was a huge risk if they had failed, their creative accounting would've landed them in hot water if they hadn't gotten promoted). Whether they stay up this season is still unknown. Basically football teams are rarely great sources of profits - they can generate profits, but are often money losing or at least high risk. That's also why though they are great for "sportswashing" - those with terrible reputations but deep pockets can invest vasts money on sports teams not for a return of profit but for reputation boosts and loyalty only sports fanaticism can deliver**.

*Then there are teams that deliberately bounce back to a lower division if they've just been promoted. They risk they ire of fans, but if you don't spend money trying to stay in the Premier League/La Liga/etc ... you collect the revenue from being in the upper league for a season, get a parachute payment from being relegated, and then have even better finances to build a better squad long term. Again comes with risks, but less so than trying to spend you didn't have trying to stay in a league you can't actually afford to be in yet. Ipswich who were last in the premier league 22 years ago are trying this tactic I believe. I think they are the ones who admitted that they got promoted to the EPL before their plan had "scheduled" and so are going to get some experience, not spend any money, get relegated, and come back. I think that's them, but maybe it was another of the recently promoted team.

**I thought to make a joke about Intel getting a sovereign wealth fund to invest in them for "techwashing" - hey works for Musk! (or at least it used to) - but of course it would just be a joke since we've already established that Intel is too valuable to the US' interests for that.

 

[Artemis]

I totally acknowledge that. I’m straightforwardly telling you they were done, and you can’t dial this in to the health of 18A. Even if 18A were a TSMC N3P analogue for mobile it would be met with starter midrange orders before Qualcomm is going to throw a flagship 8 Elite smartphone chip on it. Gelsinger can be seen more of as a long term figure in the way East Asian industrialists would have been in their early days albeit for reform, but that isn’t the system we live in and so there’s only so much federal spigot $.

There might be more now that USG senses a threat to the fabs tbf, but I am literally acknowledging to you I don’t think the entire Intel plan was coherent from a pure private sector perspective on reasonable timescales, 18A being C+ or A+ is irrelevant ultimately. F+ okay, yeah sure that weighs.




This would be my guess. The board wanted a turnaround and Intel can BS about the cash even tho fabs are bad for 0-10 year $$$$ milking X86, but it’s too hard to stomach that vs stomaching an ousting of PG when that collapses, and the writing has been on the wall.

The other thing is Intel has failed to get into AI and probably isn’t focusing enough on that.

FWIW, a lot of this (including AI) still predates Gelsinger and every report I’ve seen suggests he was doing a good job of reforming internal culture. I’m not the only one who thinks this was kind of doomed though due to the legacy.

I really believe the only way someone was going to make it through in this:

A) the Wall Street maxi plan: cut the fabs and go as hard as possible on frugal TSMC X86 chips for “milking it” money. They still decline but in silence.
B) Gelsinger, but with much more govt subsidies.

C) Time Machine and undo Intel mistakes in the 10’s.

I think it seemed as if I was defending Intel per see but ultimately by analysis was extremely pessimistic, Dave and I were just talking past one another initially.

I mean, I am not pessimistic about the fabs in principle.

But the financial and economic mudpit that Intel 2000’s and Intel 2010’s have sown are terrible and poorly timed. Suspect Wall Street also overestimated how much juice X86 has. Designing bloated and middling energy high performance CPUs while doing nothing else good and not even laterally integrating into mobile has a pretty horrible outlook for the next 0-15 years on way too many fronts.

But it’s strictly true they could make more money by driving it home of course with TSMC. It’s just, even then, I suspect they overestimated how much it has at a structural level which is probably why Gelsinger’s plan even seemed plausible.

That or they thought more government money was coming but Intel’s only getting maybe ~$15B +-5 iirc from CHIPS.

Wow, half that — just: $7.86B for Arizona and Ohio in total. That is pathetic on our behalf as a nation tbh. Bad priorities.

 

[dada_dave]

I think it seemed as if I was defending Intel per see but ultimately by analysis was extremely pessimistic, Dave and I were just talking past one another initially.

I mean, I am not pessimistic about the fabs in principle.

But the financial and economic mudpit that Intel 2000’s and Intel 2010’s have sown are terrible and poorly timed. Suspect Wall Street also overestimated how much juice X86 has. Designing bloated and middling energy high performance CPUs while doing nothing else good and not even laterally integrating into mobile has a pretty horrible outlook for the next 0-15 years on way too many fronts.

But it’s strictly true they could make more money by driving it home of course with TSMC. It’s just, even then, I suspect they overestimated how much it has at a structural level which is probably why Gelsinger’s plan even seemed plausible.

That or they thought more government money was coming but Intel’s only getting maybe ~$15B +-5 iirc from CHIPS.

Wow, half that — just: $7.86B for Arizona and Ohio in total. That is pathetic on our behalf as a nation tbh. Bad priorities.

To be honest, if Intel was forced to spin off its fabs, I wouldn't mind seeing the fabs getting more money to make sure they stayed solvent or even profitable, but otherwise you also risk simply giving Intel itself more money to hang itself with ... moral hazard and all that stuff good capitalists love to opine on in when it is giving money to poor people. For Intel, we rag on their chip design part but those are the divisions that are still currently profitable and should have to stand on their own two feet as it were. Further, those aren't the part that are of a national security priority like the fabs are. That's why I'm a little hesitant to just give them even more money in the current situation. I dunno maybe it could be structured in such a way that wasn't or was less of an issue (and that was probably a reason behind the delays and reductions in addition to the other money the government was giving them).

 

[Artemis]

Well the money is contingent to building the fabs and building those and then taking subsidies is still a net losing proposition without using them and selling to others so it works fine. I agree if we handed the “intel” today 20B annually for a while or something and let them cheat us with like only one fab that’s good and does 35K WSPM by 2030 yes lol that’s crazy.

 

[Artemis]

But in my more maximalist subsidy scenario, @dada_dave is right you can’t not spin off the fabs because it might make it economical to cheat agreements and do the bare minimum, effectively subsidizing Intel design. Whereas if IFS standalone decided to save some money after an agreement and half-ass something for a better time, as long as it’s not egregious, who cares, it’s a pure play fab & packaging firm. Eventually it’s going to semi stuff.

 

[Artemis]

Ideal is low but critical stipulations like spinning off the fabs and meeting certain buildout guidelines, but no silliness like “guarantee workers child care and comply with NEPA”.

That and a consistent dose of money one way or multiple. Guaranteed orders and/or USG effectively subsidizing Amazon, Microsoft, Google, and on the other side Apple Qualcomm Nvidia and AMD to take some orders with Intel at prices they just couldn’t refuse for midrange and budget stuff + guaranteed orders from USG itself (see defense) are a good way to do this.

Provides volume and some real constraints and liability to customers, but gives them some massive windfalls.

 

[throAU]

Yeah it makes literally no sense for Nvidia to go adopt X86 or buy a license. They have much more control with an Arm license and I suspect even Qualcomm with the recent legal battle isn’t upset enough to sincerely believe Intel & AMD are more trustworthy.

I was thinking more along the lines of Nvidia owning all of the IP to have full access to internal documentation for making better software emulation, but if this isn't relevant then ... don't need it I guess.

I agree the intel foundry is the most important element of the business, that and their patent warchest (which may also be of use to Nvidia - I'm sure intel own a lot more IP than just the IP for x86) but my confidence in them turning the foundry around based on the past 10 years of false promises and under-delivery is.... small.

 

[Nycturne]

I was thinking more along the lines of Nvidia owning all of the IP to have full access to internal documentation for making better software emulation, but if this isn't relevant then ... don't need it I guess.

If you are going for translation, you want the translation to spit out code beneficial to your own chosen micro architecture. Trying to mimic the micro architecture rather than the ISA quirks themselves likely makes things harder, not easier. And you don’t need Intel for the x64 ISA spec.

An older example: you could emulate x86 on the G5 just fine. But it required doing an endian swap every time you wrote/read memory so that little endian behaviors of the ISA were preserved. But the G3/G4 chips could operate in little endian mode which would effectively allow little endian code to operate on a system that was big endian without changes by doing some trickery with memory addresses. This had the effect of making x86 emulation faster on earlier PowerPC chips with this capability. It’s the same when we talk about flag behaviors, etc. It’s not so much about the internals of the chip we want to translate from, but rather the ISA quirks we need to preserve to avoid big performance pitfalls by having to account for them in code we emit from the translation vs letting the hardware provide the behaviors.

I agree the intel foundry is the most important element of the business, that and their patent warchest (which may also be of use to Nvidia - I'm sure intel own a lot more IP than just the IP for x86) but my confidence in them turning the foundry around based on the past 10 years of false promises and under-delivery is.... small.

The question is, how much of that patent war chest is still relevant? Are modern Intel patents the sort of tech ARM/RISC-V chip designers even want?

 

[Cmaier]

The question is, how much of that patent war chest is still relevant? Are modern Intel patents the sort of tech ARM/RISC-V chip designers even want?

their most valuable patents may be fab-related.

 

[Artemis]

their most valuable patents may be fab-related.

Exactly.

 

[Artemis]

Rumoured "abysmal 10% yield" for Intel 18A is fake news - OC3D

Intel's 10% yield rate rumours for its 18A lithography node are fake news - Here's why you shouldn't believe this news.

overclock3d.net

Well well well. Who could have predicted