Intel tapes out chips on 1.8nm and 2nm production nodes | Hacker News Reader

Intel tapes out chips on 1.8nm and 2nm production nodes

tomshardware.com

146 pointscainxinth164 comments

Since "nm" is meaningless these days, what's the transistor count/mm2 of this 1.8nm Intel node?

As reference: TSMC 3nm is ~290 million transistors/mm2 (MTr/mm2).

             IBM      TSMC   Intel   Samsung
  22nm                       16.50  
  16nm/14nm          28.88   44.67   33.32
  10nm               52.51  100.76   51.82
  7nm                91.20  237.18   95.08
  5nm               171.30    
  3nm               292.21    
  2nm        333.33

https://news.ycombinator.com/item?id=27063034

https://www.techradar.com/news/ibm-unveils-worlds-first-2nm-...

240m on 7nm node vs 290 on 3nm? Oo

Intel renamed their nodes to align better to the foundries. 10nm is now Intel 7, 7nm is now Intel 4, which would broadly look more aligned in that table if it were updated.

What does nm measure these days? I hear it's just a marketing term at this point, but "1.8" seems like a specific enough number that it must be measuring something.

Nothing anymore.

Originally it measured a specific dimension on the transistor. Then transistors started shrinking non-uniformly, so they adjusted it -- "how big the transistors would be at equivalent density if they had shrunk evenly". That's an easy thing to play with though, and so you see more and more departures from reality.

It's like RHA on tank armor.

took me a second to find the meaning of RHA https://en.m.wikipedia.org/wiki/Rolled_homogeneous_armour

I can't decide if that's an excellent analogy or a terrible one. On the one hand it's very true, on the other hand I suspect the fraction of HN readers who are familiar with tank armor is fairly small.

For those not familiar, RHA is the type of armor used circa world-war II. Armor has gotten better on a per-inch-thickness with improved technology. Modern tanks (and modern anti-tank munitions) are often rated as to the equivalent thickness of RHA protection (or penetration) they provide.

For example, Explosive Reactive Armor (ERA) can have a very high RHA rating if you use a traditional HEAT (High-Explosive Anti-Tank) round as your benchmark, but a tandem HEAT round can almost completely eliminate the per-thickness protection advantage that ERA offers. So a tandem HEAT round might have a lower RHA penetration rating than a traditional HEAT round, yet be more effective against an ERA with a very high RHA rating.

So, if you have a spiffy new tandem HEAT round, and want to fake^H^H^H^H present good numbers, you find the highest RHA rated ERA armor that it can penetrate and claim that as the RHA rating of your projectile. Even though it might not be able to penetrate that many inches of actual RHA, or possibly even modern composite armor with such an RHA rating.

Which is to say something called RHA was the metric used for tank armour measuring, but it got gamed more and more and is now unreliable on the face of it except for rough jists?

Do you think that marketing will drop to pico or zero point whatever nano?

Minimum feature size.

The problem is that we've reached a point where not all features scale down evenly. So while some transistors might be 2 nm on a new process, others are largely unchanged from 3 or 5 nm.

How much you get out of a new process is now highly dependent on how you design your chip and what you want it to do.

This is why AMD is using multiple processes in a single CPU package these days. Some elements of their processors get little or no benefit from smaller nodes, so it's cheaper to make them on an older node.

I believe it doesn’t even measure that now. It’s pure marketing numbers, albeit industry wide instead of the fault of any one company.

>What does nm measure these days?

Not Much.

It refers to the theoretical dimensions that a planar transistor would have to achieve the same density. Nobody uses planar transistors these days. Instead, 3d multi channel transistors are being used. The number is supposedly referring to the gate length of the theoretical planar transistor. If your transistors can reuse the same gate to drive three different channels, that means the effective gate length would be one third but once you open this can of worms, the numbers can be chosen arbitrarily to mean whatever you want.

It's like knots. No one's dropping a knotted rope into the water to measure speed, but the term persists.

Am I reading this correctly that at 7nm intel had >2x the transistor density of TSMC and their 7nm node is still ahead of TSMCs 5nm node on that measure? That‘s not what I‘d have expected given everything one hears abt. intel —- have they fallen off mainly by not progressing beyond that node, or along other dimensions?

Isn't what Intel now sells as "7" a rebranded 10nm process and actual 7nm which you're referring to does not yet exist commercially?

Intel 7nm on that chart is what has been rebranded Intel 4, ie the process that still isn't in high volume production. Meanwhile TSMC is in high volume for N3 (the next iPhone chips are currently using up that capacity).

They have fallen off, but not by nearly as much as the node names indicate. TSMC has been "cheating" with their node designations for years (c.f. 3nm is not remotely a (5/3)^2==2.8x increase in density vs. 5nm, it's more like 30-50% denser), where Intel stuck with the notionally numeric scales for longer.

But Intel has given up now too, the "7nm" line in that chart is actually showing the process they're marketing as "Intel 4" (denoting that it sits roughly where a "4nm" process would vs. their competitors).

The Athlon XP naming approach. Intel following AMD again! /s

Nitpick: this practice actually began way before the Athlon XP, with the Am5x86: https://en.wikipedia.org/wiki/Performance_Rating

Even the density figures are mostly lies. For example TSMCs N5 is closer to 134MTr/mm2 in the real world[1]. It can also depend on the chip design itself. Some designs may scale better than others because not all transistors get the same density improvement from new nodes. Ultimately simple metrics for predicting performance are sort of breaking down at the scale these chips are currently being manufactured at. For example, their TSMCs N3 increases transistor density for logic transistors but SRAM sees no improvement over N5[2].

[1]https://www.angstronomics.com/p/the-truth-of-tsmc-5nm

[2]https://www.tomshardware.com/news/no-sram-scaling-implies-on...

Is this some image of how the devices you'd build with SRAM (ie. big uniform array of bitcells) are very regular? I feel like logic being not-as-uniform makes the talk about density scaling different, but I can't exactly articulate how

It’s in large part due to the wire density… a 6T SRAM bit cell by itself could scale transistor density well, but the word lines and bit lines are the limiters as they typically have to go up 4 metal layers. Also, each read/write port you add to the SRAM macro increases the size exponentially rather than the linear scaling of the bit cells.

I see! Thanks for finishing the thought!

If it's meaningless then why does Apple spend a shitload of money on it? While the nm number doesn't translate to a feature anymore, the nm number is not meaningless. It is more an estimate of behavior. They behave as 2nm transistors would behave in terms of spec.

Because TSMC is the fab that can mass produce today the highest density transistor count.

The customer doesn't care about density they care about performance. If you could somehow build a magical low density but performant chip that's what people would buy.

Density = bigger cache, more cores.

Also lower electrical resistance per compute cycle (energy, heat).

Is that true? Density also means your chips are smaller which means you’re getting more parts on a wafer (though yield is important there).

This is correct from a logical point of view but not a practical point of view. With a few major exceptions transistors basically are one shape and do one thing. So if you can scale the process node you scale the density, if you scale the density you scale the performance. It’s not like 1 transistor could you give you 0.01 flops and another could give you 0.02 flops, a transistor does 1 thing and it works or doesn’t (with a few caveats), and the second order effects don’t really overcome that.

Oh, transistors are not one shape across nodes. Not anymore, and not for a long time. The planar version taught in beginning EE classes has evolved into various complicated 3D shapes.

Yeah I know, I was putting caveats all over the place because first order, things are simple, but there's lots of complication. But in reality density == performance, and transistors are mainly just interchangeable. Yes, you could theoretically get a less dense transistor with a fantastic slew rate or some such property, but it's very unlikely it'll matter in a conversation about digital logic.

> They behave as 2nm transistors would behave in terms of spec.

And how do 2nm transistors behave?

TSMC's 3nm process is still more dense than their 5nm process, just not by as much as the name would suggest. The name is meaningless because it no longer gives you an accurate idea of the average feature size.

I feel like many threads said Apple’s main advantage was process.

So with a better process Apple and mobile chips can become even more power efficient than intel chips?

Hard to piece this info together with that.

Intel never shipped any 7nm, they stayed on their 10nm which is as you can see far behind the 5nm TSMC node that Apple used. So Apple benefiting from a far superior process is real.

Intel now sells to foundry customers.

Apple is the best paying foundry customer. If Intel can deliver, Apple will use Intel fabs.

Intel has been troubled for about 10 years now. The cracks started showing in 2013 when they delayed the 14nm process for about a year, then widened when they couldn’t get 10nm into production for several years. At this point we are allowed to be a little skeptical when they tell us they are going to regain the crown.

I definitely agree with you. Though AMD's comeback with Ryzen really surprised me and I don't think other companies can't replicate comebacks like that.

After the Bulldozer fiasco who would think AMD would be where they are today?

> After the Bulldozer fiasco who would think AMD would be where they are today?

I mean... for people who remember AMD's original breakthrough with the Athlon, it's not super surprising.

Although it feels a bit like a problem with Intel as well. I think, too much time wearing the crown can be bad for a company. Although some companies don't seem to suffer from that problem - Nvidia and TSMC being 2 pretty good examples.

TSMC hasn't been on top nearly as long as Intel was.

NVidia indeed seems an outlier, although arguably the GPU division of AMD hasn't had the same overhaul that the CPU side had.

Yeah, the last significant stumble of Nvidia I can remember is the GeForce FX (aka GeForce 5) series being notably inferior to the ATI R300 series.

I do, Bulldozer was a great lesson on how to run a great team to make a mediocre product because you don't have the right feedback in place. When you have silicon is not the time to find out you perform poorly.

>I think, too much time wearing the crown can be bad for a company.

Being #1 is bad for everyone and everything period. History has shown us time and time again that the #1 spot inevitably falls from grace, the only question being how disasterously. There is no exception.

#2 and below are where you want to place if your primary interest is continued, ideally permanent, long-term prosperity.

>I don't think other companies can't replicate comebacks like that.

AMD didn't really do so much. It was TSMC that saved them.

Couldn't be a more false statement.

AMD's success was also due to their new Zen architecture and move to chiplets. TSMC alone wouln't have been enough to save them if their architecture would have sucked.

Right. It was just a coincidence that they became competitive only after selling glofo and moving to TSMC.

Please get your facts straight. Zen architecture and their move to chaplets were also big factors of their success.

AMD still used Glo-Fo for the first gen Ryzens and still do for the I/O die in some chiplets because it's cost competitive.

It they would have just ported their failed Bulldozer architecture to TSMC, that wouldn't have helped them at all.

Your original claim that "AMD didn't do much" and it was "TSMC that saved them" is a blatantly false. You are grossly misinformed if you believe that.

This isn't really true. Zen was built for years at Global Foundries on 14nm and out performed Intel at the time.

Moving to TSMC definitely helped later, but it was the original Zen architecture that put them back in the game.

https://en.m.wikipedia.org/wiki/Zen_(first_generation)

Zen didn't really outperform Intel in single thread until Zen 3. Zen 2 was a close call, though

Class-for-class, I'm of the understanding Intel still wins the single thread competition.

This didn't really matter though.

The CPU was competitive with Intel in real world scenarios and was very well received when it launched in 2017; eg: https://www.pcworld.com/article/412262/ryzen-review-amd-is-b...

The point is that the architecture and CPU design was the thing, not the later switch to TSMC which didn't happen until Zen 2 in 2019.

Ha, you don't have to convince me anything. I have several desktops and laptops equipped with various zen 1-3 CPUs. I am no intel fanboi. Architecture is important but needs manufacturing

A bit apples and oranges, isn't it? Intel has been having trouble fabbing. AMD was having trouble designing. Keller et al designed a hell of a chip with Ryzen, then threw it to TSMC to produce. Intel probably could have done the same, but for the sake of having a competitive home grown fab, I'm glad they didn't.

AMD also had trouble fabbing. They didnt spin out their fabs until 2009 and continued to use manufacturing from the spun out entity (Global Foundries) for many years after that despite their capabilities being behind Intel and TSMC. Ryzen manufacturing didnt move to TSMC until 2019 with Zen 2.

IIRC, AMD going with TSMC happened around the time GloFo said they're not going to do cutting edge anymore.

Makes me wonder exactly what happened there. Did GloFo hit trouble and give up? Did AMD tell them they weren't going to use them anymore, leading to them giving up? Basically, who dumped who?

I think AMD dumped GF. The first two Zen line ups (1000/2000) clearly used an interior process.

People were skeptical of Apple for a long time after Steve Jobs came back, too.

Intels issue is that it’s competing on technology. Jobs never really tried to compete on technology. Apple competed on user experience and its “unfair advantage” was vertical integration.

Apple didn’t become competitive on a technological basis until 2012 with the iPhone 5s.

What’s Intel’s “unfair advantage”.

Wait, isn't Intel vertically integrated exactly like Apple? They own design and foundry. TSMC is mostly foundry. AMD is mostly design.

Right now, vertical integration is a weakness for Intel -- just as it was for Apple during the Jobs interregnum and the beginning of his return. (e.g. the pre-Apple-on-Intel silicon years.)

But the situation could quickly change and vertical integration could become an advantage again for Intel.

Especially in a "challenging" geopolitical environment.

And Apple is also a bit player in each of its markets. The iPhone only has 13% market share worldwide and the Mac has around 10%.

It happens to have the most profitable niches and it can ignore markets it doesn’t want to serve.

The PC market is not growing. The chance of Intel overtaking Nvidia for GPUs, ARM for low power chips and making “old” chips for cars, microwaves, etc. is slim

> Apple competed on user experience and its “unfair advantage” was vertical integration.

The thing Apple does isn't really vertical integration. It's tying.

The only thing stopping iPhone hardware from running Android is Apple. The only thing stopping iOS from running on Samsung hardware is Apple. It's not that they fundamentally require each other. Maybe some minor proportion of iOS features require some iPhone-specific hardware. If you offered iOS on Samsung or Android on iPhone without those features, they would still sell. So why are they not offering combinations customers want?

Because if they sold iOS for Samsung, they'd lose the margin on the hardware. If they sold iPhones with Android, they'd lose the margin on the OS/services. So if you want one you have to take both.

That's not really a competitive advantage. It doesn't gain market share, it costs it, because some of the people who would buy iOS on Samsung or Android on iPhone will go buy Android on Samsung. What it does is increase their margins at the expense of consumer choice.

And it's a risk, because it means they have to execute well everywhere, or the thing that underperforms brings down the whole stack. Right before Mac switched to Intel, the uncompetitiveness of PowerPC processors was hurting them.

Which is the same thing that happened to Intel when their process fell behind, because their designers were stuck with it. So Intel learned their lesson and now Arc is on TSMC and Intel is starting to fab for third parties.

Which means all they need for success is to get back their process advantage. They may or may not be able to do that, but it's all they need.

A product is always a combination of features. That’s kind of the definition of a product. Do you also think Tesla should sell its battery to Ford? Is it “tying” when Nintendo doesn’t let you run “PlaystationOS” on a Nintendo Switch?

When Jobs came back and introduced the iMac, Apple wasn’t exactly a behemoth in 1997. Enough users chose using their own free will to buy an integrated product (with a shitty operating system) to get Apple back to profitability.

Even in 2001, when Apple introduced the iPod, Apple was going back and forth between profitability and loss.

By 2003, enough users found value in the integration between iPods and iTunes and the Music Store to choose a more expensive integrated product.

ios would be a worse user experience (on worse hardware) if it was running on Samsung phones. Users have a choice, they can buy an Android device with slower hardware, less updates, and a none integrated ecosystem for less money. 87% of users in the world make that choice.

> Which is the same thing that happened to Intel when their process fell behind, because their designers were stuck with it. So Intel learned their lesson and now Arc is on TSMC and Intel is starting to fab for third parties.

Wow, all they have to do is getter better, I’m sure they never thought about that…

And the third party they are fabbing for is MediaTek and it’s based on a 22nm process - which was stated of the art in 2012.

> Right before Mac switched to Intel, the uncompetitiveness of PowerPC processors was hurting them.

By 2011, if Apple had completely killed the Mac, it would have still been more profitable than any computer company. Fast forward to today, and the Mac only makes up about 10% of their revenue.

> A product is always a combination of features. That’s kind of the definition of a product.

So if your product is flour or plate glass... ?

Many products are combinations of things. A simple bench is lumber and screws and paint. But lumber and screws and paint are all widely available commodities. Someone who sells benches doesn't have to sell screws for them to be available in the market. They may not even manufacture the screws, so anyone who wants them can buy them from the same place they did.

When it becomes tying is when there is any part of your product that customers might want to buy separately, but nobody else makes it, so it isn't available separately, from anyone. You can only get it tied to the other parts. That's tying.

> Do you also think Tesla should sell its battery to Ford?

I think Telsa would be happy to sell batteries to Ford. They'd build another battery factory and make a lot of money.

Do you think Tesla should be able to prohibit anyone from taking the battery out of a Tesla and using it in some other car or for some other purpose?

> Is it “tying” when Nintendo doesn’t let you run “PlaystationOS” on a Nintendo Switch?

Obviously. What makes you think that is some kind of controversial result? Anybody who wants to should be able to install iOS on their XBOX.

> When Jobs came back and introduced the iMac, Apple wasn’t exactly a behemoth in 1997. Enough users chose using their own free will to buy an integrated product (with a shitty operating system) to get Apple back to profitability.

Buying an integrated product isn't the same thing as buying a product because it's integrated. By definition if they want any of the parts that isn't available separately, they have to buy the whole thing.

And as shitty as MacOS Classic was, it wasn't as shitty as Windows ME.

> ios would be a worse user experience (on worse hardware) if it was running on Samsung phones.

Not compared to the same Samsung phone running Android, presumably.

> Wow, all they have to do is getter better, I’m sure they never thought about that...

You asked what their competitive advantage would be. For years it was superior process. It's not physically impossible for them to do that again. We just don't know if they will yet.

> And the third party they are fabbing for is MediaTek and it’s based on a 22nm process - which was stated of the art in 2012.

Because their problem was caused by their existing process being a mess. They have to be able to do state of the art in volume before they can sell it to somebody else.

The more significant development was Arc on TSMC.

> By 2011, if Apple had completely killed the Mac, it would have still been more profitable than any computer company. Fast forward to today, and the Mac only makes up about 10% of their revenue.

How does that affect the state of their integrated product just before the Intel switch when they didn't have a competitive PowerPC processor?

> But lumber and screws and paint are all widely available commodities

And phones using ARM chips are sold by hundreds of manufactures with different feature sets.

> Do you think Tesla should be able to prohibit anyone from taking the battery out of a Tesla and using it in some other car or for some other purpose?

If you can take an iOS image you download to your computer (and you can with iTunes) and make it work without Apple custom hardware go for it. Or do you think that Apple should be forced to support third party hardware?

What if I want to run Teslas infotainment on my Honda civic? Should Tesla be forced to do it?

> Buying an integrated product isn't the same thing as buying a product because it's integrated. By definition if they want any of the parts that isn't available separately, they have to buy the whole thing.

So again should Apple have been forced to make the iMac modular enough so you could rip out the PPC and put in an x86 chip so you could run Windows because you preferred Windows, but you like the Bondi Blue iMac?

Should Apple be forced to sell their custom chips to other manufacturers? What if I just want the T1 security chip in my Android Blu R1 HD I bought from Amazon, should Apple be forced to sell it to you separately?

> And as shitty as MacOS Classic was, it wasn't as shitty as Windows ME

Did you ever deal with manually allocating memory for applications? The cooperative multitasking that caused all applications to stop when you clicked on a menu?

> How does that affect the state of their integrated product just before the Intel switch when they didn't have a competitive PowerPC processor?

The PPC was behind x86 from the time Jobs came back until 2005, yet it still sold well enough to represent half of Apple’s revenue.

> And phones using ARM chips are sold by hundreds of manufactures with different feature sets.

You were just making the argument that the hardware from Samsung et al isn't as good.

> If you can take an iOS image you download to your computer (and you can with iTunes) and make it work without Apple custom hardware go for it. Or do you think that Apple should be forced to support third party hardware?

It's not a question of forcing them to.

Suppose Samsung was willing to do the work to get iOS running on their devices. Apple not only won't help, they'll sue them, to keep the OS tied to their own hardware.

> What if I want to run Teslas infotainment on my Honda civic? Should Tesla be forced to do it?

They shouldn't interfere with your attempt.

> Should Apple be forced to sell their custom chips to other manufacturers? What if I just want the T1 security chip in my Android Blu R1 HD I bought from Amazon, should Apple be forced to sell it to you separately?

Why not? If they're the only supplier of something then they should sell it to anyone who wants to buy it.

> Did you ever deal with manually allocating memory for applications? The cooperative multitasking that caused all applications to stop when you clicked on a menu?

Did you ever deal with the infamous instability of DOS-based Windows, complete with a total lack of security as the biggest target for viruses because of the larger user base?

> The PPC was behind x86 from the time Jobs came back until 2005, yet it still sold well enough to represent half of Apple’s revenue.

2005 was pre-iPhone, so of course it was a large percentage of their revenue. But their total revenue in 2004 was lower than it was in 1996 -- and in 1996 there wasn't even an iPod.

For about 20 years I had the mantra of "Don't bet against Intel". A good example of this was in the early 90's when all the hype from the Apple fans about PowerPC and the end of x86 by the year 2000. The hype was real but Intel was just quietly saying "we have a road map and we will go well beyond what people expect" and then Intel did it - while PPC slowly faded away (for various reasons).

Everytime someone predicted the end of Intel - they just kept knocking out of the park when least expected. Thinking going from Pentuim 4 Prescott to Intel Core series for instance. Side note - At least the Prescott was vaguely ambitious!

But the last decade has not been great for Intel. That engineering prowess has faded a little. Their new GPU stuff is neat and very forward thinking but it is still a wait and see on that. But when it comes to their other elements, it feels like they have just been doing the same thing with very minor tweaks for far too long now.

---

Another aside, it is same theoretical greatness doesn't always translate to market greatness. If it worked like that we would probably all be using PowerPC machines in some degree. A wonderful ISA!

I visited some scientific VLSI/EDA conferences some 20 years ago. Intel was everywhere, both in sponsoring as in real participation. Apple was literally nowhere. I wonder if this has changed.

Idk about conferences but Apple sponsored some architecture courses when I was at Berkeley, and I ended up interning on an Apple hardware team due to connections made at a big hiring event they hosted on campus.

What year?

Think it had more to do with Intel partnering with Microsoft and Windows taking over the PC market in the 90s.

>Their new GPU stuff is neat and very forward thinking but it is still a wait and see on that.

That ball's in our (the consumers') court, actually. They are priced aggressively, about what video cards cost before the mining and now "AI" craze. They are competently capable, software teething pains notwithstanding.

So will Intel ARC succeed? We as consumers now have to answer. I can't tell which way the arrow is leaning right now, though software devs don't appear interested in accomodating them like Nvidia and AMD.

It is hard to tell. If Intel sticks with it and keeps the price low while still aggressively moving towards future needs - they could become a decent 3rd player.

That they have no DX11 or below legacy combined with some interesting performance profiles especially at the higher resolution areas leads me to think they are looking at where the market needs will be in 2-3 year from now. It seems like higher resolutions doesn't hamper performance are much as one would really expect. It is still way behind Nvidia and AMD but there is a lot of potential there.

I have seen a few people on youtube who benchmark cards who have made the wise choice to not touch these yet and wait for the second generation. That way they can come in when things are more stable and not have a cloud hanging over their head.

If Intel is to treat it like Microsoft did with Xbox and give it a few years to mature - it could be huge. They could also just drop it in a year or two and it just becomes a technical curiosity that we will look back on like "Remember when Inter did that?". Sort of like the i960 processors. Something really neat but a dead end for them.

I remember than in 2017 there were rumors that Intel was developing a novel microarchitecture from scratch to beat AMD and it should be out in 2020, did any of that pan out?

People really overstate Intel's woes. They got stuck on 14nm for ages, but were for all that time still the market leader. AMD has never comprehensively challenged them; only one segment at time is an AMD product better.

This article about Intel candy canes and unicorns and rainbows looks like simply a large hit of cocaine to keep their bloodied investors hopes' up!

To quote Newt from alien 2, "Ripley, it won't make a difference!" because SRAM is NOT scaling down!

https://fuse.wikichip.org/news/7343/iedm-2022-did-we-just-wi...

And? Just because one aspect isn’t scaling well doesn’t mean the rest being scaled is pointless.

Besides the fact they will probably figure out a way to either replace sram with something else that does scale better or they will figure out novel techniques to scale sram.

HBM is scaling up!

Isn't this problem solved with a chiplet design like AMD does? Just manufacture the SRAM using an older and cheaper node and keep the logic in the most advanced node.

Eventually you kick all SRAM out of the compute die and the Levels start meaning Layers.

That's what they did on their GPUs with HBM.

HBM was a bit of the opposite. That was off-chip DRAM being brought very close to the GPU.

As that article points out, CPUs consist of both logic and RAM. Even in a worst case scenario where RAM doesn't scale down you still get a huge performance and/or efficiency boost.

To quote:

> Of course, the impact won’t be felt the same across the board. The percentage of SRAM and caches on chips varies greatly by your target market and overall capabilities

and

> Beyond SRAM, the industry has been looking into many other alternative memory architectures. Emerging memory technologies include MRAM, FeRAM, NRAM, RRAM, STT-RAM, PCM, and others. Those emerging memory bitcells offer unique tradeoffs when compared to SRAM such as higher density at lower read/write specifications, non-volatility capabilities, lower read-write cycle capabilities, or lower power at potentially lower density or speeds. While they are not direct replacements for SRAM, moving forward they might play a role as level 4 or level 5 caches where the lower performance tradeoffs can be offset by higher density.

This seems so good (almost too good) that I wonder if they are being realistic. It seems like Intel has recently been missing all their projected timelines.

Considering they are talking about full-on manufacturing in about a year, they should be pretty far along. But I can't help being a little worried that it will happen.

To be fair, they started missing all the timelines after Brian Krzanich succesfully executed his political coup as CEO of the company.

It's hard to overstate how toxic he's been to the company.

Is there context to this claim? If so, can you share it?

The performance of Intel during the five years that he ran it is plenty of context, no? 2013 is a pretty clear inflection point for Intel and that was the year that he became CEO.

I arrived at Google search in 2013 and soon they were complaining about how much Intel was "Slacking" and how badly they needed an alternate supplier for data center chips.

I followed Chromebook and laptop iGPU 3D performance very carefully from 2013-2019. Did you know Intel made ZERO iGPU performance improvements in this time period? The 2013 IRIS Pro 5200 and Haswell (4000 series) were a major highpoint for FIVE years. The 5000 Through 8000 Intel core cpus were highly stagnant and forgettable! Mostly they were slacking off with new designs like the Ynnn "thermal throttlemaker" Such chips ran fast for 3 secs then quit - completely! From 2013-2019 their marketing pinheads sure had a field day making up new names for the same old shit iGPUs year after year after year!

I have seen NO impact from Pat Gelsinger so far! Almost-competitive CPU's using 250% of the power of Ryzen 7000? Are you kidding us, Pat?

Kryzanich laid off a huge number of CPU architects in 2014 so they could have a much bigger and of course - more important marketing department! Who needs these useless digital designers when all they can do is reduce the mips per watt? I mean who cares about that? And while we're at it why not staff the boardroom of the world's greatest physics company with community college marketing people? Because, why not?

I don't see Apple returning to use the Intel i9-thighburner CPUs anytime soon!

What's interesting is if you put all of that next to the stock price. The disconnect is so weird.

Short term gains traded for long term losses is a pretty standard CEO playbook move. Make yours then bail with a golden parachute as the plane crashes.

I think he was kicked out.

With no severance?

That's a rhetorical question?

I recall that he was entitled to double digit millions compensation if he resigned (which makes no sense to me so maybe I'm remembering this wrong). I'm sure there is an article somewhere on the web with all of the details and whether he actually got any of it or not but I would assume that he did. If only to make it easier for Intel to find its next CEO.

Gelsinger's only been there 2 years. You don't expect a major turn-around for roughly 5 (at the least) since it takes a pretty long time to go from research node to production.

> from 2013-2019. Did you know Intel made ZERO iGPU performance improvements in this time period?

This is absolutely not true at all. Haswell was a major perf-per-watt improvement compared to previous Ivy Bridge. Broadwell and then Skylake both had significant improvements over their successors, not only in performance but also in power consumption and by adding necessary GPU tech to them. Then both Ice Lake and Tiger Lake had near 2x improvements in some industry standard benchmarks (e.g., Manhattan).

The difference between Intel's 2019 and 2013 integrated GPUs is much bigger than the difference between their 2019 and 2013 processors.

Not sure what he's referring to, but during his tenure he decided to "shake things up" and laid people off almost-Elon-Musk-style. He basically had a target in terms of Dollars, and he asked HR to come up with metrics to hit the target. They laid off everyone who had had a bad review the prior year. But that wasn't enough to hit the target, so they laid off everyone who had an average performance review, but low bonuses. But that didn't hit the target, so they offered voluntary severance packages to those who had not had a promotion more than N years. That, along with encouraging early retirement let them hit the target.

The layoffs happened all at once - not in phases. You had cases where a manager was suddenly told to inform his report that he is being laid off, only to then be told that the manager himself is being laid off.

As you can imagine, any time someone does a blunt top down approach like this, you lose a lot of critical people with no one in the team knowledgeable to take over.

No one saw it coming. Managers were not told in advance.

You also lose a lot of fucks from people who know things but it isn't their job title either.

After big cuts everyone is covering their asses.

As toxic as Brian Krazinich was, you can’t really blame one person for the problem at Intel. Yes, he and Murthy initiated the layoffs. But then the orders were executed by mid/senior management. They are the ones who have used reviews/promos/bonuses as political weapons for as long as I ever worked there (over a decade). It was natural that the 2016 layoffs at Intel eviscerated the technical contributors while keeping the politicians, ass-kissers and powerpointers in place.

As far as I can tell Pat Gelsinger has not changed anything. The same hacks who ruined Intel are all still there, moving up the managerial ranks. Makes sense because Pat came from that culture… he practically defined it.

> Considering they are talking about full-on manufacturing in about a year, they should be pretty far along

When they say "manufacturing ready" in 2024, that doesnt mean you will be able to purchase anything built on those nodes next year. Intel 4 was manufacturing ready last year, but final chips being available isnt expected until the end of this year (and availability of final chips is usually months before you can buy, for example, a laptop with one in it).

If these milestones aren't verifiable maybe we shouldn't be reporting and commenting on them.

Its of course not possible to know what will happen in the future, but the particular topic this article is about was announced by Intel during their Q322 earnings release:

"Intel 20A, 18A: First test chips have taped out"

Quote from page 5 of the Earnings Presentation here: https://www.intc.com/news-events/ir-calendar/detail/20221027...

Note that the product of "tape-out" is just data. That data is then used to produce masks, which are then used in photolithography to define features on a chip.

Now, I could go off and tape out some test circuits for a 1pm process. That doesn't mean I'm making any progress towards bringing a 1pm fab node to reality.

Would be great to get more competition from Intel in the fab space.

Would also be great if intel would stay in the GPU race as a third player to put some pressure to AMD and Nvidia.

> Would also be great if intel would stay in the GPU race as a third player to put some pressure to AMD and Nvidia.

Are they not?

Their discrete GPU business has had loses over 3.5 Billions so far and it will likely keep loosing money for the near future since they aren't competitive yet and the next gen, Battlemage is not planned till 2024 the earliest.

Their ability to support the losses of the GPU division to keep them in the race for the long run will depend on the other businesses making bank to compensate.

Are they making bank? If not, they might axe the GPU division.

But I wouldn't hesitate to buy one if it's the right card for your needs. I'm likely going to. Even if they axe the discrete cards the tech will make it into CPU's and SOC's so driver support will continue.

And of course their Linux support is open source and top notch.

Yeah, as many other things as I dislike about intel, I give them some grace for how well they support their GPUs on linux. If I could only figure out how to get flux.jl to work with oneapi.jl

I’m loving my A770. Absolutely zero issues with it and it’s handling raytracing and all with aplomb.

I recall seeing some issues with performance on some older games - do you know if this is still the case, or did you not try it? I'm not in the market for a graphics card at the moment, but if there's another competitive company in the space when I am that would be excellent.

I think those massive performance issues were only restricted to games using DX9 because the architecture had no support for it in HW so what they did was use the translation layer DX9 to Vulkan for those titles and performance went up. Shouldn't be an issue anymore.

Ah yeah I remember DX9 support was mentioned. That's good to hear though

The latest driver claims to improve DX9 performance by 60%. Counter-Strike by 120%.

Myself I don't care about old games. As long as they're "good enough", I worry about unreleased games that will force me to get a new video card before I would otherwise.

Nobody has good support for DX9 in their hardware, it's all in the drivers. And that's why the delta between Nvidia and everybody else is so much higher for DX9-DX11 than it is for DX12, Metal & Vulcan.

Tbh I haven’t had a chance to game a ton yet in it but everything I’ve tried has been fine.

Buuut I only have a 2K monitor and the system is also massively overkill (i9-13900k @6.4ghz, 64GB DDR5 @ 6400mhz, striped PCIe4 NVMe drives, Optane PCIe OS drive etc etc). So I’ll be hard pressed to really notice any performance issues since I upgraded from an i5-10500 with 16gb ddr4 and a Radeon Pro WX5100 8GB lol.

other perspective - in the long run, intel can't afford to axe the gpu division, because there's so much potential in that space where x86 and its new generations is increasingly getting bigger competition from arm based designs.

Intel basically has to get gpus right, eventually. AMD is already doing it. Apple is already doing it, for their purposes. Intel has to catch up.

>intel can't afford to axe the gpu division, because there's so much potential in that space

Haha, that's adorable when you consider how many things with potential Intel has either axed, sold off, or abandoned over the years.

  - ARM mobile chips (StrongARM)

  - Mobile modems (now belonging to Apple)

  - Compute stick

  - Itanium

  - The Edison and Curie

  - IoT Gateway

  - Xeon Phy

  - Movidius VPUs haven't been axed but they don't seem to be getting developed further

  - Recon smart glasses

None of those were core to Intel's business in the way GPUs are likely going to be going forward. The closer integration of CPUs and GPUs is an industry wide trend with Apple, Nvidia, AMD, and Intel all trying to eliminate the need to go off-chip for as much as possible. Things can be dramatically faster and more efficient across both consumer and server workloads if everything is just part of the same actual chip and you don't have to go out and use things which are plugged into the motherboard. Making that work is a lot easier if you have both your own CPU and GPU architectures. Apple has both in the M1, Nvidia is pairing ARM cores with its GPUs, AMD obviously makes both, and Intel is trying to get in on the game. They recognize that having their own GPU designs are going to be very important to both their consumer and HPC products going forward, especially as use-cases which rely on GPU or GPU-like acceleration (AI stuff) proliferate. Having competitive GPU designs is critical to their future competitiveness.

> ARM

Oh what could have been

Intel actually spent too much effort on the Itanium, it was stillborn. The Xeon Phi was just the product name of the Larrabee without the display output parts, and the Larrabee was demoted to the Xeon Phi BECAUSE it didn't have enough potential to compete with the discrete GPUs from Nvidia and AMD. The current attempt at discrete GPUs is more promising.

Also, they helped develop Infiniband before dropping out. Much later they got back into the Infiniband business by buying QLogic, and from there developed Omnipath which they eventually gave up on, spinning the product line off as a separate company, Cornelis Networks.

Some others: Realsense, Optane, and they sold their NAND busisess to SK Hynix.

Don't forget the Intel purchase of McAfee software for $7.5B

I think they just junked their network switch silicon. Again.

But Intel replaced the Xeon Phi with their discrete GPUs, I would abandon those too, if I had an even better product to sell.

But if they can't afford to keep it running then they will have no choice but to axe it.

They went into this knowing they'd spend billions before making money. Its taken almost a decade of development and Alchemist is out on the market and the drivers continue to improve. Intel's a big company with 28 billion in cash on hand, I don't see them giving up in this space when AI/ML is becoming a burgeoning cash cow.

>They went into this knowing they'd spend billions before making money.

They went into this during the covid and crypto driven GPU scalpocalipse when everything that could render triangles would sell for whatever sticker price you put on it but ended up launching right after the bubble bust. Very unfortunate timing. Now their products need to actually be competitive.

Intel didn't just jump into the gpu market because of the 2020 mining craze. It takes much longer to develop these products. They hired Raja Koduri in 2017:

https://arstechnica.com/gadgets/2017/11/intel-poaches-amds-t...

>They hired Raja Koduri in 2017

Ah yes, the person who led the development of the least successful GPUs in Radeon's history. Intel looked at his achievements and thought "that's the rockstar we need".

Thank God I find another person on HN with similar views on Intel's GPU and Raja.

They need to enter the compute market. From what I can tell Intel is the only company serious about OpenCL and even they decided to abandon SVM (shared virtual memory) from OpenCL 2.x and instead introduced their own unified shared memory extension (USM). USM is basically how SVM should have worked in the first place and can be easily implemented on any platform, even non Intel GPUs. Meanwhile SVM is almost impossible to get right and only AMD and Intel ever implemented it and Intel iGPUs already share system RAM with the GPU so there was actually nothing to implement.

Exciting news. 1.8nm in late 2024 is quite soon.

Intel to Apple: If you can't accept me at my worst you don't deserve me at my best.

>"Intel to Apple: If you can't accept me at my worst you don't deserve me at my best."

Thank heavens for that. Good riddance.

Yeah, I can't imagine wanting to work with Apple.

Me neither, with all those hundreds of millions of devices sold annually.

Outbidding every other company for the fab.. it’s truly a nightmare

Looked at from the fab’s viewpoint, that is just selling their capacity to the highest paying customer. Seems pretty much standard business practice. Nothing nefarious.

That post was using the fab's viewpoint. Nightmare was sarcastic.

Intel Foundry Services (https://www.intel.com/content/www/us/en/foundry/intel-foundr...) needs customers.

We’ll see whether they turn down Apple when it offers to buy stuff worth billions.

Also: how large are Intel’s nanometers, compared to those of the competition? I wouldn’t know.

I know you're kidding but this is capitalism, not a spouse relationship.

If they pay, they’re always welcome.

Pat even said they wanted Apple's business back in an interview.

Of course. If Apple brings the money Intel will gladly fab their chips.

Not many people are aware, but for a while Intel was manufacturing parts for the iPhone. They made no money out of it - Apple demanded too low a price.

I don’t think they did. Apple made an offer and they refused. One of Intel’s worse mistakes.

https://www.wired.com/2016/09/oh-hey-intels-iphone-now/

Thanks

It wasn’t so obvious a few years ago. You couldn’t have Intel fab your design for any price. x86 was as strategic for them as their fabing expertise.

Now it’s a different story.

Intel's fabless offering has been around for many years. It's historically been uncompetitive and uninteresting. Arguably it still is. Their ability as a fabless partner is not the same as their ability to manufacturing their own highly custom designs.

Intel did say that they were hoping to win Apple's business back (https://www.macrumors.com/2021/10/17/intel-hopes-to-win-back...). I doubt that Apple will go back to x64 architecture, but Intel is offering to work as a third-party foundry for others' chips and Apple is willing to pay a ton more than most companies (probably including Intel's own internal accounting) for the latest fabrication processes.

If Apple is willing to pay such that the margin on Apple's business is better than the margin on Intel's own chips, I don't think Intel is going to be vindictive. In fact, Apple choosing Intel for their fabrication would probably be the best thing for Intel's fabrication business. Not only can Apple pay lots of money in advance, it would signal to the industry that Intel's fabrication process is something they should all be paying attention to. Apple tends to be very risk-averse with their supply chain and Apple is known as the company that pays for the best fabrication. If they move to Intel's fabrication, they've done a lot of due diligence in that process. Yes, Intel could offer a good deal to Apple to get them to move so some of it might be marketing - but that's exactly what Intel could use for its third-party fab business.

I doubt that Intel will hold a grudge. It would be way more valuable to Intel to win Apple's business. Heck, TSMC is on the cutting edge today in part because Apple will commit so much money. If Intel can win Apple's business, they can remove a lot of the wind behind TSMC's sails. Yes, TSMC will still be a great company, but Apple's willingness to pay premium prices definitely helps TSMC. They know that investing in new, expensive processes will be paid back by Apple. If TSMC has to rely on Qualcomm and MediaTek (who want to wait for nodes to become cheaper), it's likely that they won't have the same purchasing for their latest nodes leading them to slow down their progress a bit. If TSMC's progress slows, that gives Intel so many benefits. All those ARM datacenter processors that people might move to? Not if TSMC's progress slows and Intel gains and keeps a sizable advantage on fabrication.

If Intel can snag Apple away from TSMC, it will probably be the best thing Intel could do for itself.

You said yourself that Apple is risk averse. All those "ifs" about Intel should tell you it ain't gonna happen. Apple pays TSMC a lot of money because they can deliver, not because they "have a plan".

Apple still might shift some non-leading edge node work over though. It's not good for Apple to have only one supplier. Having an alternative to TSMC is in their interest.

> Intel to Apple: If you can't accept me at my worst you don't deserve me at my best.

OR: Dear Apple, pay us 30% of your revenue or get kicked out of our FabStore.

Can’t wait to see negative Nanometers

High voltage anti-electrons that produce power rather than consume it while we're at it.

I really have no electrical engineering knowledge, so this question might be naive. Given that we are approaching a hard limit (the size of the atom), I wonder wether we could branch out into (true) 3d? So far everything is 2D and if I think of the brain I picture it as an interconnected 3d-blob, which just seems much more sensible.

Nice seeing angstroms used in real life!

This is the first time I have read the term “tape out”. Please clarify for non-semiconductor people, thanks.

How does Intel China coexist with Intel USA without proprietary information leaking out?

Doesn't Intel only produce memory in China, and is in the process of moving all of that production to SK?

Strict export controls on information. People from countries that are subject to export controls aren't allowed to see certain things or work on certain projects.

I'm not sure how big a presence Intel has in China these days, but my impression is that it's small and it shrunk quite a bit when Intel got out of the NAND memory business.

I'm curious how this is enforced within the company ecosystem. Seems that it can easily be violated.

Edit: I was going only off the headline, never opened the article. original comment below.

---

Somewhat surprised we're not measuring in Angstroms yet.

18 and 20 angstroms respectively. Might not be a SI unit but it's still metric and gets rid of that pesky decimal. Especially since a silicon atom is 2 Angstroms, it seems like the right unit.

Did you read the article? It's literally what Intel is doing.

I didn't. I was going off the comments and headline. Thanks for the call-out, it's well deserved.

Did you read the HN Guidelines [1]? It's literally saying not to do what you did:

> Please don't comment on whether someone read an article. "Did you even read the article? It mentions that" can be shortened to "The article mentions that".

[1] https://news.ycombinator.com/newsguidelines.html

Reply on news.ycombinator.com