I realize that all the process sizes aren't all necessarily equivalent ("TSMC is on 5nm already!"), but this could really be an existential problem for Intel.
Windows already has an ARM version. Apple is about to release its own silicon. Amazon is running its own silicon on servers.
It seems like peak Intel is already behind us, but just how quickly might they fall?
It seems insane to me that Intel won't have 7nm fabs working at scale until 2023 but AMD is selling $99 CPUs printed at 7nm already.
How did Intel go from 24 months ahead of everyone to 4 years behind??
The most advanced process node was arguably Intel's only true differentiator in the past. Now they make CPUs with comparable IPC to ARM or AMD but they have to use old process tech. And they are only getting further behind.
How smart was it for AMD to spin out their fabs into GlobalFoundaries? After all 7nm Ryzen is done by TSMC.
You are correct though the exact numbers don't matter. Intel is already behind and is having problems where as other manufacturers are moving to smaller nodes. Exact node size does not matter just the fact that Intel is being left behind
It's just like, back in the 2000's, Intel insisting GHz was the only measure that mattered (compared to AMD's lower clocked but more efficient designs), but in the other way around.
I wouldn't count Intel out yet. They've been leapfrogged by AMD before, and then caught up. The cycle repeats.
Also, do you remember the 90's? There was a promising 64-bit CPU called the DEC Alpha. Windows once even had an Alpha version. We know how that turned out.
The problem with Intel is that they have been missing important deadlines by massive amounts of time. Their 10nm node was slated to arrive almost 5 years ago. It is still far from mature.
TSMC is readying serial production of 5nm (equivalent to Intel 7nm) this year.
I wouldn't plan on seeing a 7nm product from Intel in any volume until 2024, and I wouldn't place money on that as a bet.
So? They have a runway, yes, but it's a freakin long one. They have a massive warchest and a crapload of market share that isn't going to go away overnight.
Id put money on them already having a new operation crush well underway by this point.
In addition, many years of using stack ranking to fire 10% or sometimes more of its employees each year, which we know from other examples like Microsoft catastrophically damages a company in many ways, including loss of many of your best employees.
Intel (and HP) countered with Itanium, which was a break from (i.e. incompatible with) the x86 architecture (IA64). It tanked.
AMD created AMD64 which was x86 compatible. They released chips like the Athlon which were very popular. This turned out to be the winning strategy.
In a strange twist, Intel was forced by the market to adopt the AMD64. It released it's own version, Intel64.
Intel messed up there too in the 90s too. It wasn't until the Core CPU roadmap that Intel regained its lead.
There seems to be a link -- Jim Keller [1] was involved in the design of DEC Alpha chips, and then moved to AMD to work on the K8 and then co-wrote the AMD64 instruction set. He also worked on Apple's chips and Tesla's HW. He recently just left Intel. What a career.
No. The announcement was the usual platitudes. We can infer from the unexpected and apparently sudden nature that there was some bust-up or dispute but nobody knows why.
The interview with Fridman may offer a bit of insight. Keller is a firm believer in Moore's Law, that it'll continue for a long time yet. He believes you have to redesign things nearly from scratch on a regular basis, to benefit from this firm progress.
Intel, meanwhile, is struggling with Moore's Law. First 10nm, now this. Backporting architectures to older nodes. It may be that Keller realised how deep the production issues went (perhaps he learned that 7nm was starting to fail too) and concluded he wasn't going to be able to achieve what he wanted to achieve when constrained by Intel's fabs.
One of the best bits I took away from that interview is Jim's comment that a lot of things need to be "rearchitected every decade, or even faster".
It's such a throwaway comment, but I think there's deep insight behind it from someone with extensive experience in the matter.
I also feel it applies to many other things, not just ASIC design.
For example, every customer I go to manages their VMware farm just like they managed their bare metal boxes in the past. Just as they're oh-so-slowly wrapping their heads around the design concepts that are a better fit for virtualisation, they're moving to the cloud. So of course, they deploy everything in the cloud just like they're used to in their on-premises VMware environment.
It's shocking how much low-hanging fruit is out there, just waiting to be rearchitected by someone with understanding and a clear vision.
As a random example: I just saw a bunch of Azure servers with empty 1TB "Application" D:\ drives. The team that built it was used to how VMware thin-provisioned drives only allocate 2MB blocks at a time, so the 1TB max capacity is just a high water mark that has no real cost. Meanwhile, Azure bills based on capacity, not usage, so that 1TB is burning money.
It seems like a small thing, but if you store 1GB in a 1TB volume in the cloud, you're not paying $123/TB/month, you're actually paying $12,288 per terabyte of data stored per month!
If you look around, it turns out that everywhere you look, people are doing "what they've always done", and it leads to crazy inefficiencies.
The reality is most places keep running software architected for an era over 2 decades ago now and it’s ossifying harder if it’s still carrying enough business inertia. Those workflows of bootstrapping VMs and EC2 instances like it’s 1996 are not going away because to do anything cloud native in your architecture you need cloud native software, and usually if you can get a container you can get an RPM or Deb and play package jockey rejecting the new technologies literally meant to do half the work for you.
In most of the cases where places just dump money it’s usually a question of labor cost spent to optimize vs the gains, and unless your business is built around scaling a lot of small customers like the usual SaaS unicorns customer acquisition is super long, painful, and technical inefficiency is the default for enterprise as a rule. It’s worth paying $200 for a $1 part because the overhead and risk of renegotiating anything is not worth it. When an hour long meeting essentially costs a minimum of $1000 to a company essentially, it’d better be worth it.
When it comes to ASIC designs and VLSI the technical debt is pretty different because each generation of hardware has past benchmarks primarily to drive it forward. Oftentimes in software people tend to want to keep things the same which discourages innovation or touching.
Wasn't it because DEC Alpha targeted the workstation market while Intel disrupted them with PCs? The story is in "Innovator's Dilemma".
Today Intel is facing the same problem DEC faced, that they make most of their money from workstations and datacenter hardware, while facing disruptive innovation that is ARM
If Clayton Christensen is correct, then Intel is dead in the water
When Alpha launched, proprietary software was the de-facto standard. Windows ran on it, but Office didn't. Neither did a huge number of popular Windows desktop applications. Alpha was reduced to a limited role in servers where the lack of software wasn't an issue and performance was important enough companies procured hardware that would be harder to remission. Alpha lived on for a good couple years running OpenVMS and Tru84 proprietary enterprise OSs and, for some lucky people, Linux.
Now most mission-critical software is one recompilation away from running on ARM. Or POWER, or RISC-V, or SPARC (for now). That AWS offers ARM-based servers is seen as a non-issue. I run the same workloads on x86, ARM and s390x and they all feel the same (except that z15 hardware is ridiculously fast). If tomorrow someone comes up with cloud servers based on RISC-V or MIPS that are competitive in price/performance with x86 and ARM, I see an easy migration path between these architectures.
In terms of mobile, that's the same. If tomorrow someone releases an Android phone running on RISC-V instead of ARM, that'll be largely irrelevant to most software.
On the desktop, Apple (and NeXT) have a couple extremely successful architecture changes under their belt. They should have no problem with this one.
Porting was difficult enough that Itanium and Alpha never got much Windows software and, therefore, were relegated to small niches, with servers and single-purpose workstations. It seems Microsoft made it right this time and that porting Windows software to ARM is easy and seamless. We'll see what happens.
> It seems like peak Intel is already behind us, but just how quickly might they fall?
Don't forget that AMD has a vested interest in keeping x86 alive as well. it's not just Intel. So it's not like x86 is stuck in the mud because of Intel's situation.
This could really be an existential problem for Intel.
Not quite, but it could be as big an event as the 737 Max debacle was for Boeing.
As I understand it, the 7nm processes in production do not use EUV, but Intel's does. Is that correct?
I'm amazed that anyone can get that "zap tin droplet with laser to emit X-rays" light source to work as a production process. Samsung claims to be in full production with it.
Apple is all about controlling all parts of their phone and software.
Now they have a well working ARM chip they have all the control over which is good enough for at least the non developer models of their laptops so it was just a question of time until they switch.
I mean why _long term_ maintain two OS'ses and two CPU archs when you can just focus one one. (And yes I'm convinced Apple will turn macOS into iOS+some desktop features in the long run).
Is there a legit case for doing it as an exercise for the experience (shaking out bugs, making discoveries by having 2 discrete development platforms), or as a strategic card, so they “have options”? I don’t know how the costs scale - whether it’s <2, ~2, or >2, and at what cost any (potential) benefits are simply cost prohibitive. Anybody have insight into this business, esp with something approaching a major player like Apple?
Yeah, they switched to Intel and went "Hey, you can run windows on it in VM or in bootcamp, maybe even play some games!". Then iPhone came along and propped the sales of macs.
Now there are people tightly locked into Apple ecosystem that doesn't exist anywhere else. Saying it as someone who left that eco-system not that long ago and boy do I miss it.
What supports that proposition? Plenty supports the opposite, such as Intel causing massive problems for several years e.g. to the MacBook Pro line, with overheating and thermal issues. They've been failing to innovate at the speed Apple would otherwise be able to, for a long time now.
> "Intel causing massive problems for several years e.g. to the MacBook Pro line, with overheating and thermal issues."
Do you not realize how ridiculous this statement is? Apple knows what CPU models are going into their laptops, they could properly design hardware and software to prevent overheating and degradation.
There are plenty of arguments, but E2E control is probably the strongest one.
It's not only about CPUs. Remember Nvidiagate or Radeongate? If Apple takes full control of the hardware like they did with iPhone/iPad/Watch they will (most likely) end up making better products.
Once Apple started making their own silicon there's no going back.
There's no good reason that so much of the world is using an architecture that has consistently shown less efficiency, especially since the rise of distributed computing. It is bound to happen.
Everyone and their dog are whooping Intel. AMD on x86, and GPUS. ARM on everything else. Modem business, flop. Nvidia is trashing them on accelerator cards and GPUs. Amazon is starting to use its own ARM CPUs and AMD. Intel really doesn't have an answer to "twice the cores, for less" value prop for cloud providers selling vCPUs.
They are still profitable, because there is a lot of remaining momentum in business markets and laptops. Those who have historically only bought intel and produced intel like a dogma take some time to change their ways. But change is happening, and it will hurt Intel badly...
Isn't there another issue in that AMD's processors may be better (for some value of better obviously) but their software and sales is not?
It's anecdotal but I don't think I've ever seen any HPC documentation or case studies go anywhere near AMD tooling (https://news.ycombinator.com/item?id=20647031 contains a discussion of this - I've never needed to do that kind of work on an AMD CPU, so I couldn't comment myself).
AMDs software and HPC docs may be inferior. But especially in HPC, benchmarks and flops/€ are the deciding factor. You don't buy a bunch of processors from intel just because they also sell you a nice compiler (with a diminishing lead in performance) and tuning tools. Thats only a secondary or tertiary factor at best.
The problem is that Intel still can't ship 10nm in mass production.
Apple's has been shipping all new chips, in TSMC 7nm, since the iPhone XS in 2018, at tens of millions per quarter, in fact nearly every product Apple sells is 7nm.
They probably sell more Watches every quarter with 7nm than Intel has sold 10nm in all time. It's a mess and TSMC will ship 5nm this quarter. They are seriously behind in MASSIVE ways.
The same tactic as with 10nm. What they say doesn't mean much. They will admit just as much as the lawyers advise them to. They will probably release some 7nm devices in homeopathic doses just to show the shareholders "see, we are delivering, you can't sue us".
At some point they probably will be ready to produce in quantities but we have no indication when that date is.
Well, now we know why Apple initiated the move to ARM this year. They must have known about this for quite some time and Intel probably shared this new timeline with them. Apple's ARM will on on 3/5nm nodes when Intel hits 7nm.
AMD is sold out due to demand and Price/Performance, Intel is sold out dude to incredibly large inability to ship anywhere near enough quantity. It's probably 5+ AMD 7nm units per every Intel 10nm.
Meanwhile Apple has been shipping 7nm for two years at probably 10 times the units per quarter.
85 comments
[ 4.1 ms ] story [ 159 ms ] threadWindows already has an ARM version. Apple is about to release its own silicon. Amazon is running its own silicon on servers.
It seems like peak Intel is already behind us, but just how quickly might they fall?
How did Intel go from 24 months ahead of everyone to 4 years behind??
The most advanced process node was arguably Intel's only true differentiator in the past. Now they make CPUs with comparable IPC to ARM or AMD but they have to use old process tech. And they are only getting further behind.
How smart was it for AMD to spin out their fabs into GlobalFoundaries? After all 7nm Ryzen is done by TSMC.
Both are brand names at this point.
[1] https://www.tomshardware.com/news/intel-says-first-10nm-desk...
Also, do you remember the 90's? There was a promising 64-bit CPU called the DEC Alpha. Windows once even had an Alpha version. We know how that turned out.
I really wish Intel was doing better, and I hope they come back, somehow. The chip world needs good competition.
TSMC is readying serial production of 5nm (equivalent to Intel 7nm) this year.
I wouldn't plan on seeing a 7nm product from Intel in any volume until 2024, and I wouldn't place money on that as a bet.
Id put money on them already having a new operation crush well underway by this point.
AdoreTV has a video on the topic: https://www.youtube.com/watch?v=agxSclh27uo
Intel (and HP) countered with Itanium, which was a break from (i.e. incompatible with) the x86 architecture (IA64). It tanked.
AMD created AMD64 which was x86 compatible. They released chips like the Athlon which were very popular. This turned out to be the winning strategy.
In a strange twist, Intel was forced by the market to adopt the AMD64. It released it's own version, Intel64.
Intel messed up there too in the 90s too. It wasn't until the Core CPU roadmap that Intel regained its lead.
There seems to be a link -- Jim Keller [1] was involved in the design of DEC Alpha chips, and then moved to AMD to work on the K8 and then co-wrote the AMD64 instruction set. He also worked on Apple's chips and Tesla's HW. He recently just left Intel. What a career.
[1] https://en.wikipedia.org/wiki/Jim_Keller_(engineer)
Well, they licensed AMD's and CALLED it their own.
The interview with Fridman may offer a bit of insight. Keller is a firm believer in Moore's Law, that it'll continue for a long time yet. He believes you have to redesign things nearly from scratch on a regular basis, to benefit from this firm progress.
Intel, meanwhile, is struggling with Moore's Law. First 10nm, now this. Backporting architectures to older nodes. It may be that Keller realised how deep the production issues went (perhaps he learned that 7nm was starting to fail too) and concluded he wasn't going to be able to achieve what he wanted to achieve when constrained by Intel's fabs.
https://youtu.be/Nb2tebYAaOA
It's such a throwaway comment, but I think there's deep insight behind it from someone with extensive experience in the matter.
I also feel it applies to many other things, not just ASIC design.
For example, every customer I go to manages their VMware farm just like they managed their bare metal boxes in the past. Just as they're oh-so-slowly wrapping their heads around the design concepts that are a better fit for virtualisation, they're moving to the cloud. So of course, they deploy everything in the cloud just like they're used to in their on-premises VMware environment.
It's shocking how much low-hanging fruit is out there, just waiting to be rearchitected by someone with understanding and a clear vision.
As a random example: I just saw a bunch of Azure servers with empty 1TB "Application" D:\ drives. The team that built it was used to how VMware thin-provisioned drives only allocate 2MB blocks at a time, so the 1TB max capacity is just a high water mark that has no real cost. Meanwhile, Azure bills based on capacity, not usage, so that 1TB is burning money.
It seems like a small thing, but if you store 1GB in a 1TB volume in the cloud, you're not paying $123/TB/month, you're actually paying $12,288 per terabyte of data stored per month!
If you look around, it turns out that everywhere you look, people are doing "what they've always done", and it leads to crazy inefficiencies.
In most of the cases where places just dump money it’s usually a question of labor cost spent to optimize vs the gains, and unless your business is built around scaling a lot of small customers like the usual SaaS unicorns customer acquisition is super long, painful, and technical inefficiency is the default for enterprise as a rule. It’s worth paying $200 for a $1 part because the overhead and risk of renegotiating anything is not worth it. When an hour long meeting essentially costs a minimum of $1000 to a company essentially, it’d better be worth it.
When it comes to ASIC designs and VLSI the technical debt is pretty different because each generation of hardware has past benchmarks primarily to drive it forward. Oftentimes in software people tend to want to keep things the same which discourages innovation or touching.
I think this is the first time they are behind on both process and chip architecture at the same time.
Today Intel is facing the same problem DEC faced, that they make most of their money from workstations and datacenter hardware, while facing disruptive innovation that is ARM
If Clayton Christensen is correct, then Intel is dead in the water
When Alpha launched, proprietary software was the de-facto standard. Windows ran on it, but Office didn't. Neither did a huge number of popular Windows desktop applications. Alpha was reduced to a limited role in servers where the lack of software wasn't an issue and performance was important enough companies procured hardware that would be harder to remission. Alpha lived on for a good couple years running OpenVMS and Tru84 proprietary enterprise OSs and, for some lucky people, Linux.
Now most mission-critical software is one recompilation away from running on ARM. Or POWER, or RISC-V, or SPARC (for now). That AWS offers ARM-based servers is seen as a non-issue. I run the same workloads on x86, ARM and s390x and they all feel the same (except that z15 hardware is ridiculously fast). If tomorrow someone comes up with cloud servers based on RISC-V or MIPS that are competitive in price/performance with x86 and ARM, I see an easy migration path between these architectures.
In terms of mobile, that's the same. If tomorrow someone releases an Android phone running on RISC-V instead of ARM, that'll be largely irrelevant to most software.
On the desktop, Apple (and NeXT) have a couple extremely successful architecture changes under their belt. They should have no problem with this one.
Porting was difficult enough that Itanium and Alpha never got much Windows software and, therefore, were relegated to small niches, with servers and single-purpose workstations. It seems Microsoft made it right this time and that porting Windows software to ARM is easy and seamless. We'll see what happens.
Don't forget that AMD has a vested interest in keeping x86 alive as well. it's not just Intel. So it's not like x86 is stuck in the mud because of Intel's situation.
Not quite, but it could be as big an event as the 737 Max debacle was for Boeing.
As I understand it, the 7nm processes in production do not use EUV, but Intel's does. Is that correct?
I'm amazed that anyone can get that "zap tin droplet with laser to emit X-rays" light source to work as a production process. Samsung claims to be in full production with it.
Totally broken Power Management on Latops (that's why I got AMD-based laptop and am very happy with this).
Problems with wifi and bluetooth chips.
Lost the anti-trust lawsuit to AMD.
Not so good performance compared to recent AMD.
And from what I heard even in industrial fields they are a lot of complaints lately.
Even Apple is leaving Intel.
Not to mention spectre/meltdown fiasco back then.
Apple is all about controlling all parts of their phone and software.
Now they have a well working ARM chip they have all the control over which is good enough for at least the non developer models of their laptops so it was just a question of time until they switch.
I mean why _long term_ maintain two OS'ses and two CPU archs when you can just focus one one. (And yes I'm convinced Apple will turn macOS into iOS+some desktop features in the long run).
https://www.youtube.com/watch?v=wgEKDGNjqsk
I'm not sure he's entirely right, but I found it pretty compelling, overall.
Now there are people tightly locked into Apple ecosystem that doesn't exist anywhere else. Saying it as someone who left that eco-system not that long ago and boy do I miss it.
Now is a good time to switch to their silicon.
Do you not realize how ridiculous this statement is? Apple knows what CPU models are going into their laptops, they could properly design hardware and software to prevent overheating and degradation.
It's not only about CPUs. Remember Nvidiagate or Radeongate? If Apple takes full control of the hardware like they did with iPhone/iPad/Watch they will (most likely) end up making better products.
But also:
- Single architecture on all their products
- More perf per watt
Once Apple started making their own silicon there's no going back.
There's no good reason that so much of the world is using an architecture that has consistently shown less efficiency, especially since the rise of distributed computing. It is bound to happen.
https://news.ycombinator.com/item?id=23934078
I haven’t heard much about it but the video makes the point that Chinese firms will be adopting it.
An open source cpu, with no licensing fees, is bound to catch on.
Could they recover from this death spiral?
It's anecdotal but I don't think I've ever seen any HPC documentation or case studies go anywhere near AMD tooling (https://news.ycombinator.com/item?id=20647031 contains a discussion of this - I've never needed to do that kind of work on an AMD CPU, so I couldn't comment myself).
Sales, dunno.
Had TSMC seen Intel's problems early, they might have been able to build more fabs to put more pressure on Intel.
Apple's has been shipping all new chips, in TSMC 7nm, since the iPhone XS in 2018, at tens of millions per quarter, in fact nearly every product Apple sells is 7nm.
They probably sell more Watches every quarter with 7nm than Intel has sold 10nm in all time. It's a mess and TSMC will ship 5nm this quarter. They are seriously behind in MASSIVE ways.
Back in March they said
> Intel CEO Bob Swann has stated that 7nm CPUs will ship in Q4 2021.
And now they are on 2022/2023...
At some point they probably will be ready to produce in quantities but we have no indication when that date is.
Only a one year delay; less than the title implies.
> Swan also said "we have root-caused the [7nm] issue and believe there are no fundamental roadblocks"
Heard that one before. Would like to believe it's true this time, but—who knows.
https://www.eetimes.com/intels-10nm-node-past-present-and-fu...
https://www.bloomberg.com/news/articles/2020-07-24/intel-con...
This means our appetite for computing power is increasing.
Meanwhile Apple has been shipping 7nm for two years at probably 10 times the units per quarter.