Hacker News Reader

Comment by threatripper | Hacker News Reader

This means that Intel manufacturing is used where it makes sense and vice versa isn't used where it doesn't make sense.

I can see those general outcomes:

* Intel manufacturing is soon again better than other manufacturers in all aspects as it has been a long time. All Intel business goes to Intel manufacturing and things return to what they were.

* Intel manufacturing is better in some aspects and continues to outcompete other manufacturers in some areas. Most Intel business goes to Intel manufacturing but some special business goes to other manufacturers. Some other companies use Intel where they have an advantage. Everybody wins.

* Intel manufacturing cannot keep up with the market leader. Most Intel business goes to the leading Fab. Intel manufacturing cannot catch up as they don't earn enough money to do so. They start cutting costs and compete downmarket with lower tier fabs. In the long run Intel becomes fabless.

View on HN

Another possibility is that the market is changing. The age old Xeon/Core/“binned low end” model may not be viable anymore as hyperscale cloud providers start to rule and people don’t care about CPUs as a differentiator. Add in COVID chaos and the answers are unclear.

Hell I just replaced thousands of older desktops with slower laptop chips — nobody noticed. Frankly, I only bought new devices because of Microsoft’s bullshit Windows 11 requirements.

The guaranteed cash flow train is slowing down, which makes setting billions of dollars on fire with high risk fab processes a big deal.

Intel had to adapt like this in the early 90s when memory became a commodity. We’ll see if they make it.

> Hell I just replaced thousands of older desktops with slower laptop chips — nobody noticed. Frankly, I only bought new devices because of Microsoft’s bullshit Windows 11 requirements.

Big enterprises think like that yeah, because they can't get their crappy Dell Inspirons to new employees anymore.

Personally I just had to get a Ryzen laptop replacement temporarily because Lenovo took too long to repair my thinkpad. And the result was delightful. Better sound, fantastic compile times, beautiful screen for half the price of my thinkpad. Next time I'll get a cheap Ryzen and if it breaks just buy a replacement device instead of relying on a pro intel enterprise device with crappy overpriced enterprise service.

Agreed - but that’s a problem for Intel, as they need enterprises to lap up those shitty Dell laptops to keep the throughput going!

NYC Public Schools buy like 200,000 computers a year. A decade ago, it was probably 50% higher due to faster refresh cycles. There are a lot more big dumb enterprise endpoints than you might think. When I sold computers in college to commercial customers, the average refresh was 30-40 months. Now it’s closer to 60.

> Hell I just replaced thousands of older desktops with slower laptop chips — nobody noticed.

This is certainly plausible if these were mostly Office users but how confident are you that you'd know if they did notice? Most of the large IT departments I've interacted with would post glowing self-evaluations while the users had plenty of complaints and/or were doing their real work on personal devices. Depending on how the business is structured and the relationships, this can be hard to measure without a fair amount of effort — I've heard people say they weren't going to report something because they didn't think it was worth the effort or they were sure that the IT people would notice and fix it.

The shift to the cloud has turned computers into glorified terminals again. For terminals CPU power is a lot less important than for a personal computer.

I understand the argument but I don't think that's enough – even on a recent Intel laptop CPU, you'll be hearing the fans if you have more than one of Outlook, Teams, etc. active or any enterprise security software even if most of the data is being stored on the cloud. Something like iOS or ChromeOS is a lot closer to delivering on that idea in my experience.

> I understand the argument but I don't think that's enough – even on a recent Intel laptop CPU, you'll be hearing the fans if you have more than one of Outlook, Teams, etc. active or any enterprise security software even if most of the data is being stored on the cloud.

Google Docs runs well on a Chromebook. As does MS Office online. The modern desktop has really turned into a terminal - it's jut using HTML & JS and works... pretty well.

Yeah, that’s why I said Office users — the core experience isn’t bad at all any more. It’s usually other apps which are pushing CPU limits notably. A big one recently has been video conferencing since a lot of schools didn’t used to need that at all, much less large classes entirely online.

Video conferencing works really well on a $200 Chromebook or $100 tablet.

Have to disagree here. Cloud apps (in a form of web apps) demand more CPU performance, not less! Apart from specialized apps that really do a lot of computing (rendering, modelling, model-training) that can easily be done in cloud, most apps are slower (or functionally constrained) if they're moved to cloud and used as web app. Think Office 365 suite - every app of the suite is either slower as web app, or functionally constrained.

For me personally, web apps only got fast when I got a computer with M1 processor.

Indeed, WordPerfect, word, excel, etc, all worked just fine on a pentium chip of the day. Yes there was always something faster but really… it was just fine.

We invested on memory. Shitty electron apps like Teams chug ram!

Yes, the market is changing -- on all fronts for Intel. Many people talk about this being a tech problem, really its more like the logic at companies when you hold a close to monopoly position -- why invest tons of R&D dollars when you can just increment more slowly and make the same money.. So they end up with a loss of tech leadership when they easily could have had it. The biggest mistake that Intel appears to have made was their failure to bring up the new 13nm EUV processes at the same rate as TSMC -- probably they were hoping it would fail, but it did not. Now TSMC has done a couple of nodes with EUV and they have some customers that are spending tons of money right now like Apple, AMD, nvidia, qualcomm, etc and now TSMC is flush with cash and have a strong runway in front of them -- its like TSMC is already in execution mode while Intel is still trying to get things working. Intel basically thought, hey lets see how long we can milk our big cloud customers with $2000 processors and only minor improvements. Those customers realized they could hire a chip team themselves and build their own processor for the amount of money they were paying Intel each year, so they did that. So they got attacked on the enterprise side by their own customers making their own chips like Amazon and Google, the consumer side by AMD just throwing lots of cores at the problem -- and Apple dumping x86 in their laptops -- and soon to be others, and they essentially have nothing in the mobile space. So they went from being dominant 10 years ago to being 2nd place in multiple of their core businesses. Clear failure of leadership to realize they need to maintain technical supremacy otherwise they will not be able to charge the prices they want to charge. Now their competitors are shipping chips with 64 cores and Intel's plan to uber slowly release 16 cores, then 20 cores, 24 cores, etc over a 8 year period or whatever their plan was is blown up..

I'm still bullish on $AMD to break much further into the enterprise space. So many AWS instances still run on Xeons. That's slowly changing, especially now with stuff like Graviton , but I think AMD can go much higher just on enterprise.

The problem that the industry faces is that the economics and reliability of these chips have been undermined on the recent nodes.

Sophie Wilson has said that cost-per-transistor is rising since we shrank below 22nm, and that a 7nm die cannot run more than half of its transistors at the same time without melting.

Sophie addresses the cost rise at 22:00.

https://m.youtube.com/watch?v=zX4ZNfvw1cw

Heat is more of a problem because finfet can't dissipate as well as planar, and reliability is declining at smaller nodes.

"With a planar device, you do not have to bother about self-heating. There are a lot of ways to dissipate heat with a planar device, but with finFETs that is not the case. The heat gets trapped and there are few chances for that heat to get dissipated."

https://semiengineering.com/chip-aging-becomes-design-proble...

https://news.ycombinator.com/item?id=29889951

If I needed a 20-year service life, I would not choose 7nm.

Edit: I did not realize that ARM1 had fewer transistors (~25k) than the 8086 (29k), and over ten times less than the 80386 (275k). Intel should have bought ARM in the 80s; instead Olivetti got them.

This problem is not specific to Intel. Many chips are not using the smallest node sizes and old fabs get a second life producing those chips instead of CPUs. That could soon be the future for Intel fabs.

That's where a fab goes to die. It needs to be fully capitalized at that point or else you don't have the money for R&D on leading edge. Intel going this direction is the direction of GloFo and no longer even attempting leading edge node production in the US anymore.

> Sophie Wilson has said that cost-per-transistor is rising since we shrank below 22nm

The data has been saying otherwise. 5nm is the only node that increased $/transistor beyond it's previous node (7nm), and that's at a time when Apple payed out the ass for a monopoly on the node except testing shuttle runs from competitors but isn't a sign of a fundamental increase in cost.

https://cdn.mos.cms.futurecdn.net/Unwdy4CoCC6A6Gn4JE38Hc-970...

> and that a 7nm die cannot run more than half of its transistors at the same time without melting.

That's true for trying to run multiGhz designs in classic plastic package BGAs like most cell phone chips, but that's been true for a while, hence why flip chip packages are a thing. Actually having a heat spreader connected to the die goes a long way.

Wilson's comments aren't incorrect from a certain point of view, but tend to get extrapolated out of her niche to the greater industry in a way that pushes the statements into inaccuracy.

Thanks, the cost per wafer does look convincing. I wonder where Wilson's figures emerged.

I think what's happening is that initially a node is more expensive, because of consolidation of the market and supply/demand amongst all of the fabless customers until capacity fully comes up. Once we're back into steady state we see the traditional economics of $/transistor falling.

That sort of coincides with TSMC having competitive, close to leading edge nodes (so the 28nm timeframe) which would line up with the rumor. The information simply hasn't been updated over the timeframe of the node. Previous to that the cost of the node was pretty fixed as long as someone like ARM cared about, now there's a lot more economic effects from the increased buyer competition that heavily changes final cost over time.

I believe her talk was from late 2017, so 7nm would have been expensive.

At the same time, AFAIK Intel was doing quite well at 14nm finfet even then (likely better than any other foundry?), but that production capacity was not available to ARM, so I guess it didn't count.

Yeah exactly. I want to be clear, I've got a tremendous amount of respect for Sophie Wilson; she's a much better engineer and more connected to how the industry is going. Her statements simply require a lot more caveats than they are normally given. It's more about the much the changing position of ARM and TSMC in the market place than anything else.

> At the same time, AFAIK Intel was doing quite well at 14nm finfet even then (likely better than any other foundry?), but that production capacity was not available to ARM, so I guess it didn't count.

Yeah, and Intel was right in the middle of their biggest misstep. Intel 10nm was in risk production for Cannon Lake with awful yields and therefore a huge $/transistor. It got shipped anyway as one SKU in May of 2018 that didn't make financial sense (there's rumors that management bonuses were tied to that release), before being relegated to process improvements for years until just recently.

It would have been fair for her to extrapolate a trend there that actually ended up being more complex in hindsight.

Anecdotally, I recently had a CPU fail for the first time and it was a 7nm one. Sent it to AMD, they verified the failure and sent a new one back. Meanwhile I have had assorted 22nm/14nm processors around the house chugging along for years without any issues.

We're going to need stuff like processors with built-in microfluidic cooling layers. Why not have club sandwich construction with every other layer working on heat extraction, power, or both? I see a future with cube-shaped processors with hundreds of layers.

I think there is a pretty clear outcome, Intel will go fabless for highest end chips and continue as normal with everything else. They will have many customers (USG) who will require domestically manufactured chips. But they also need to compete with AMD, Apple, and friends and this will allow them to fab out to TSMC while, for lack of a better term, saving face.

This wont work. Moving high margin parts to TSMC will wither Intel fab resources to keep up in the mid range. Soon enough you start to struggle like GlobalFoundries.

Fabless scenario will not happen, ever.

One simple reason that similarly to aeronautics chip manufactures are critical industries for US.

US government will not let either one die.

We don't need small nodes for military/aero chips. It would suffice to keep some old fabs running to secure supply. The US needs fabs but they don't need Intel as a company.

Splitting out the fab business and letting it run into the red however could incentivize the government to save the fabs and the jobs with billions of bailout money while the other Intel continues to rake in money with TSMC chips.

This assumes that US military has no need for servers, desktops, or mobile devices.

Which will be covered by fabs built on US soil which presumably are contractually obliged to manufacture chips for the military on demand. (And obliged to take the precautions to be able to do so.) Some of them can be (ex) Intel fabs, some TSMC or Samsung. I don't see any hard necessity for Intel to keep their fabs in house when there are other solutions.

Increasingly we do. The military is very much embracing edge compute currently.

Literally every sentence is out of touch and outright wrong. I don't even know where to start.

> We don't need small nodes for military/aero chips. It would suffice to keep some old fabs running to secure supply.

What? How do you think the military works, via smoke signals? They need computer, a LOT of computers to operate day to day. And that PCs for just boring clerk work. Not to mention the idea that old chips would somehow be good enough to keep up technologically, both military and economically.

> The US needs fabs but they don't need Intel as a company.

Oh yeah? They will just a walk a engineer corps to a fab and have them press big red manufacture button few million times to make enough chips right? Can't be more difficult than that...

What is the problem with an Intel design manufactured in a TSMC fab built on US soil? There will be fabs on US soil in the future but they might not be owned by Intel.

It probably depends on whether the chip is a commodity chip or something special

https://en.wikipedia.org/wiki/International_Traffic_in_Arms_...

hmmm I dunno... maybe China taking Taiwan and banning all export of electronics to US, leaving US with pants down

If China invades Taiwan, the rumor is that Taiwan's strategy is to scuttle everything at the last second so China isn't additionally incentivized to invade just to get the electronics manufacturing infrastructure.

So there won't be any electronics manufacturing to ban at all, but it leaves the US in the same place.

The military doesn't just need chips for cruise missiles and radars. There has actually been a big push the last 20 years to move a lot of systems to COTS architectures. An Arleigh Burke destroyer uses Xeon powered blade servers to run its Aegis combat system not some weird proprietary chip from the 80s.

The level of computerization and networking is going up in the military so the needs will only increase. Intel's CONUS fabs are a national security concern.

Does an Arizona fab of TSMC count as an American fab?

> US government will not let either one die

This isn’t Intel’s shareholders’ problem. Split the businesses. If the U.S. wants to bail out Intel’s fabs, let them. (Though a better strategy would be a general subsidy for anyone building fabs in America, which should be a thing.)

Well, what if TSMC created an independent US based manufacturing presence that was DOD qualified? Or one of the other currently non-US manufacturers?

I am not certain, but I think they can’t. I think those critical functions cannot have foreign owners or officers if they want to sell to the government.

Agreed except there's only 2 outcomes: #1 (win) or #3 (lose). The chips business is a winner-take-all market where if your product isn't the best (in efficiency, performance, whatever metric), then it can't command a price premium nor the attention of OEMs. Also the core business isn't consumer but for enterprise/datacenter market where efficiency/performance is paramount.

If #1, then everything is rosy and Intel will regain its dominance from the days of Sandy Bridge. Its chips will be fabricated internally and they get to enjoy the profits from vertical integration.

If #3, then Intel will certainly spin off or sell off its manufacturing/fabrication business very similar to AMD more than a decade ago.

The caveat is that I don't think Intel's problem right now is simply a manufacturing issue. Sure their Alder Lake is competitive but it's not superior to AMD's offerings or even measured against Apple's SOCs. Remember that unlike the last iteration (Rocket Lake), Alder Lake doesn't suffer from a mismatch in design vs. manufacturing cadence - it's arrived as expected.

Reply on news.ycombinator.com