89 comments

[ 4.8 ms ] story [ 195 ms ] thread
Intel is supposed to be releasing their Xe-HPG this year and they've already announced it was going to be manufactured externally. Is this supposed to be specifically about the CPUs?
I have really mixed feelings on this. I'd prefer to see Intel work with TSMC and/or Samsung to build/operate joint fabs or some kind of co-ownership over completely outsourcing. I'd prefer to see a few more diverse production locations that include an EU fab, and updated US fab locations. I'm concerned about instability with China stretching its' elbows in the region and I think having more geodiverse locations would be better for everyone involved.

It's kind of the end of an era and sad to see in a lot of ways. I do want to see Intel be more competitive as they'll probably be treading water in mindshare for the next couple years.

Aside: I want to know how much Intel is paying laptop mfgs to keep higher end memory and GPU options out of AMD powered laptops.

Your concern about China is Taiwan’s leverage with TSMC to remain independent and pro Western.
Could you expand what you mean in your comment (just a little bit) by any chance?

I am very interested in this line-of-reasoning.

TSMC guarantees that every western country has a stake in Taiwan's sovereignity.

As long as they manufacture most of the worlds advanced silicon, western countries will continue to protect and arm taiwan.

Well EU just recently announced some multizillion EU fab plan
(comment deleted)
I wouldn't take that plan seriously. It probably wil be a huge waste of money, and yield no results.
Yet another Bloomberg "bombshell" without a single named source?

Rumour were going on for at least 3 years, but they are rumours.

Except for the fact of Bob Swan being sighted visiting Xinzhu, there was no other facts supporting that whatsoever in a those years.

How is this a bombshell? They have already announced some of their new GPUs are going to be manufactured externally and due for release this year.
Yes, GPUs, not CPUs.
What exactly is bombshell about replacing G with C? If anything it was bombshell when it happened for GPUs originally planned to be made in house, it cascading to CPUs in the same time frame would almost be expected.
Giving up manufacturing would be terrible. It really feels like Intel is ceding any hope of leadership to focus on milking their IP for cash. That would be incredibly frustrating and sad.

I hope it's not true or just very limited scope.

It's worked out extremely well for AMD... and recall everyone was saying it would be the end of AMD when they spun off their FAB into Global Foundries.

Fabricating, and design are very different businesses.

Well yes, but while AMD has great technology, their success as a CPU maker has always relied on the success of the x86 platform.

The assumption is x86 will remain competitive for years to come, but that seems a lot less clear now. It feels like what ARM is doing right now is quite similar to what x86 did to mini computers and mainframes.

ARM has become a competitive pressure that will prevent Intel and AMD from cashing in on that platform the way they have in the past. Chromebooks and the iPad have been nibbling at the low end for a few years now. Now Apple's M1 is taking a big chunk out of Intel's lucrative ultrabook market. Microsoft and other OEMs are going to start pushing harder on ARM just to stay competitive.

Though being the sole licensee of x86, maybe AMD can pull a weird rabbit out of their hat.

AMD also has ARM irons in the fire[1].

[1] https://www.amd.com/en/amd-opteron-a1100

I didn't realize AMD had ARM CPUs.

I was trying to hint that AMD might make some weird hybrid x86/ ARM beast that could have the advantages of ARM much of the time, but could run x86 native. I'm not sure that's even possible.

AMD might do well in a post x86 world. It's just a broader playing field.

That would indeed be an interesting setup, kind of like the big.LITTLE configuration popular on a lot of high-end ARM processors.

I'm sure it could be done... but probably would have some interesting effects on software (unless you do full x86 emulation on the ARM cores... but then what's the point).

It's far more likely we'll start seeing ARM processors with hundreds or thousands of cores in the server market first. These would be particularly interesting to virtual server providers like Amazon, Google, Digital Ocean and others.

AMD stopped building these in 2016 and Amazon hired away most of the AMD team that worked on this chip.
This was several years ago. They never really launched their ARM line. I believe they no longer have an ARM license. They certainly dont have a product ARM line
Extremely well? The split happened 12 years ago - AMD was struggling almost bankrupt until last year or two. Global Foundries is also not doing so well - they also gave up on chasing the latest process and focus on specialized 12nm stuff
Extremely well, yes. AMD, in fact, did not go bankrupt.

Keeping their Fab arm was quite literally going to bankrupt them back when they made the decision to spin it off.

They could not afford to sink the R&D cost into newer, smaller process fabing and recognized maintaining their own Fab would (a) bankrupt AMD attempting to push into newer processes and/or (b) prevent AMD from ever being competitive in the x86 market due to using old process.

In hindsight, it was a brilliant move, and now has allowed AMD to partner with any Fab that has the technology AMD needs.

It was brilliant -- but it reduced the number of leading-edge fab players from 3 to 2. Intel pulling the same move would reduce the number of leading-edge fab players from 2 to 1.

See the problem?

What? Intel hasn't been a "leading-edge" fab in years and years at this point.

And that is exactly why they've lost their "advantage" over AMD. Intel got bogged down in very expensive R&D, and failed over and over to produce yield good enough to mass produce retail chips.

As far as I know, TSMC and Samsung are both considered leading-edge.

Samsung is burning money to produce chips a node behind TSMC. They are a leading-edge fab right now like Intel is a leading-edge fab right now. You're right that they are a contender, but the fab market is always going to be this way: one player will be the best, the others won't be. If you design CPUs but don't manufacture them, your supplier will always have a natural monopoly, meanwhile your customers have more options every day. Ditching the monopoly market to focus on the competitive market is not a good long-term strategy.
Intel has been stuck on the same process since at least 2018, and continually delayed their next node process.

Right now current estimates put Intel's 7nm node at least 2-3 years away... and that's after all the delays that they have already experienced. It's so bad, Intel is seriously considering outsourcing chip production for not just GPU's, but CPU's as well - that means Intel believes it's in a really bad position.

By the time Intel finally gets their 7nm process working at acceptable yields... 7nm will be old news. TSMC is working on 5 and 4nm processes right now, and of course AMD will be using them to their advantage at the earliest date possible.

Meanwhile, Samsung shipped AMD chips for Ryzen 1 (and 2?), and is producing GPU chips for AMD to this day. TSMC is producing Ryzen 3 (and possibly other things for AMD). Both companies will be producing something for Intel soon, it seems.

Like it or not, Samsung is obviously producing things Intel cannot.

Clearly, allowing Intel to use current state-of-the-art fab processes from whichever company at the time happens to be the best is a sure winner for Intel.

I'm well aware of the situation, and no, 2018 was not the beginning of the problems.

> allowing Intel to use current state-of-the-art fab processes from whichever company at the time happens to be the best is a sure winner

That's the type of one-move-ahead thinking I expect investors will love right up until AMD and Intel give up all their margins bidding against each other for the kingmaking process. Intel would still be behind 50% of the time, when they lose the bidding war, with the difference being that the 50% of the time they are ahead they don't get to keep the margins.

I'm curious what you feel Intel should do then?

Intel's deep pockets have got to be starting to feel lighter at this point - and they're looking at many more years of sinking money into a black hole before being able to produce useful things. (during which time AMD will be leapfrogging to newer fab processes as they become available)

We're already starting to see OEM's use AMD cpu's like we haven't in nearly 2 decades. The server and datacenter markets are coming around to AMD and it's Threadrippers as well. Eventually Intel's sweetheart OEM deals will expire, and who knows if they'll be renegotiated or ditched completely? That will be exceptionally painful for Intel... as those OEM deals are largely what Intel has been coasting on for all this time.

I just don't see an option for Intel, unless they want to gamble their competitiveness possibly permanently. They're not quite in dire straights like AMD was when it spun off Global Foundries... but the outlook is much the same.

> That's the type of one-move-ahead thinking I expect investors will love right up until AMD and Intel give up all their margins bidding against each other for the kingmaking process.

Seems like a major win for consumers. There's no reason to have a $10,000 consumer CPU today, with the exception of lack of competition in both price and performance.

> In hindsight, it was a brilliant move, and now has allowed AMD to partner with any Fab that has the technology AMD needs.

For AMD, it was the best possible move both long and short term because they never had a leadership position. That doesn't mean it's the best thing for Intel. Intel has for years commanded a premium because of their branding and platform power. They need to demonstrate a technological edge or their margins are going to collapse.

There's a reason that every desktop and server CPU (ignoring APUs) AMD is shipping have a 12nm die manufactured by Global Foundries in addition to the 7nm CPU chiplets. It's a brilliant workaround within the scope of their wafer supply agreement with GF.
amd buying radeon is really what almost bankrupted them. they barely got both divisions working properly now although gpu is behind by about 1 gen on some stuff.
GPU is now on par minus the wank RT stuff, stuff you have to disable anyway if you want to have good framerate.
>AMD was struggling almost bankrupt until last year or two

a year or two ago AMD use Global foundries. a year or two ago they switched to TSMC. AMD should ditch Global foundries a long time ago.

AMD had volume requirements through Global Foundries as part of the spin off. This ensured Global Foundries would have initial volume to prop it up until it was viable for other customers.
I think this is the wrong way of looking at it - and actually the way that Intel themselves describe it. Intel think of themselves as a manufacturing company whose designs drive demand for their fab. But in truth, most of their value comes from x86 which could be produced by any fab. To stay competitive, their high end designs need to be on the absolute bleedge edge process. Which up until now has been Intel fabs.

But it's not giving up on Intel fabs to get their stuff produced by TSMC. They need to do both - they need to make money by producing the best designs in the market no matter what the process in the high end space, as wl as sinking investment into their fabrication side to make sure that in 5 years time they're the leading fab. At which point they can turn around and say F-U to all those TSMC customers, produce only Intel chips and use that to capture market share.

And the way I see it that's probably the reason why TSMC won't allocate any significant capacity to them, they already have more demand than they can produce.

Maybe if they spun out the fab part but they have so much stuff under the corporation that probably makes no sense.

AMD spun out global foundries, and they're not competitive at the bleeding edge today. I fear the fab-only part of Intel would be the same.
So then AMD switched to TSMC. Otherwise they would have been stuck with an inferior process and fallen even more behind during the last decade.
> But in truth, most of their value comes from x86 which could be produced by any fab.

This is why I suggest it's a profit taking move. If they ditched all manufacturing, their profitability will certainly go up in the near term.

But then how does Intel fend off increasingly competitive ARM CPUs? So long as Intel had a hand in manufacturing, they at least have a chance of maintaining an edge through their processes.

I'm also deeply frustrated that yet another manufacturing giant is leaving the US. But that's more or less another issue.

But it is good "profit taking": selling good x86 products that can't be made with Intel fabrication processes (which apparently offer a choice between too old, not working or not ready) instead of selling nothing (at least, not in satisfactory quantities).
> But it's not giving up on Intel fabs to get their stuff produced by TSMC.

Manufacturing at the leading edge is getting increasingly expensive, to the point where every fab other than TSMC, Samsung, and Intel has bowed out over the years and has never operated on the leading edge again.

A major part of the reason for this consolidation is that fabs need more and more economies of scale in order to make up the cost of the initial investment in the fab. If you can't achieve those economies of scale, then the manufacturing node will never be profitable, and whoever's bankrolling it will quickly pull the plug. (GlobalFoundries exiting the 7nm market is the most recent example of this.)

If Intel outsources leading edge production to TSMC or Samsung, they will lose operational experience and infrastructure needed to operate fabs at the leading edge, and their fabs will never be competitive for leading edge production again (barring a major misstep by TSMC or Samsung).

It might be a short term move just to sell, until they reintegrate in-house manufacturing. It would avoid fab being a comparison spot. But it would also devalue the brand to the eyes of consumers potentially.
Intel has whiffed their 10nm process node to the point where it may never launch in any scale. Their 7nm node is significantly behind schedule. Their only solution in the near-term is to keep increasing the TDP of their chips, which won't work for long.

This might be their only good option. Sure outsourcing manufacturing is a risk, but it's probably one they need to take. They will catch back up if they can get their 7nm node back on track, which isn't a given.

I just think it stinks of desperation and suspect once they go down this path it's going to be very difficult to reverse course.
Stink or otherwise, they are actually very desperate!
Intel is intel because of x86, not because of their fab. 10nm is what, 4 years behind, does anyone feel confident 7nm will be ready in the next 5 years? Having a superior architecture (which they don't anymore) can only compensate for so much regarding node size. Intel isn't going to pull that one around. Having your fab in-house only makes sense if you're state of the art. At some point, not admitting defeat and just keep increasing TDP is false pride.
Intel is what it is because of fab. For many years, they were way ahead. 5 years ago AMD also had x86 but was very much behind and almost went bankrupt.
The federal government has identified semiconductor manufacturing as a critical piece of national infrastructure. If it comes to it, Intel can get subsidized like Boeing.
Is money actually the issue here?

I presume it was incorrect technical decisions rather than a lack of money.

More money can fix a lot of mistakes, like making it financially viable to start from "scratch" (whatever that means in context), or fund multiple redundant teams trying more ideas in parallel.
Gotta fucking love it. Our American industries are becoming so incompetent and corrupt that we just start giving them free cash in hopes that they fix themselves. Instead it just becomes permanent socialism for corporations and by socialism, I mean free hand outs to the non-engineers MBA fuckheads that run these companies to drain them for cash.
>socialism for corporations and by socialism

That's not how Socialism works, but you can call it elitist and extreme capitalism if you like...or United Cyberpunk of America.

Can someone explain why intel's fabs are in so much trouble? I tought everybody is using ASML, so what are the important differences?
The ASML lithography machine is just one, very important, machine out of many dozens. A semiconductor fab process involves 100s-1000s of process steps each with dozens to hundreds of parameters. For example, something as simple as rinsing a wafer after an etch step:

- How long to rinse

- What to rinse with (DI water, solvent, etc)

- What temperature to rinse at

- Should the temperature change over time?

- Agitated vs. non-agitated

It may sound small, but a bad rinse process will tank yields and make the whole process fail.

In a huge semiconductor company like Intel or TSMC, multiple process engineers specialize on each one of these steps out of hundreds. These are truly massive undertakings.

> It really feels like Intel is ceding any hope of leadership to focus on milking their IP for cash.

AMD/Apple/Nvidia gave up manufacturing took all the cash and left Intel a long way behind. Its manufacturing is causing their problems and losing leadership.

I don't think Apple/ Nvidia have ever manufactured their own CPUs before. Apple has certainly been fabless their entire run as CPU makers.

> Its manufacturing is causing their problems and losing leadership.

Certainly. But not long ago they were top dog. Ceasing manufacturing their own CPUs would mean they could never be more than a peer among equals. I suppose that's a step up from their current state, but they can kiss their 30% margins goodbye.

Semiconductors are in massive demand right now. If Intel can offload some production, it will surely use the freed-up capacity to produce other chips.
Well, in the U.S. and Europe people who actually produce something are hindered by bureaucrats and ideologues.

It is no wonder that valuable knowledge moves to Asia.

Funny, ARM, AMD, and Apple seem to be having no issues. Maybe they should change the company name to Antel?
To be fair to OP all of these companies outsourced production to Asia.
ARM takes it further and just licenses its IP, leaving to its licensees the decision of whether to outsource production to Asia.

Not all necessarily outsource to Asia. I think STMicro and Infineon still produce some ARM microcontrollers (microcontrollers; these are not smartphone CPUs) in Europe.

Not exactly.. ARM doesn't really produce their own chips and there are manufacturers that use ARM cores that produce chips in the EU (ST, Infinion and NXP as well I think) and US (TI). AMD still has GlobalFoundries produce things for them (and they have fabs in the US and in Germany).
Or you know, look at the pattern. All the companies that are succeeding are run by engineers at head, be it AMD, ARM, Apple, Google, Microsoft (soon to be going full speed into their own chip designs on the down-low because qualcomm is sputtering), etc.

The companies like Boeing and Intel are failing because MBAs with no engineering backgrounds completed their coups in the last decade and have purged any engineering promotions in their orgs. They are entirely 100% profit milking companies now.

Conspiracy theory: Intel is using their deeper pockets to buy up TSMC fab capacity in an effort to slow AMD production. Doing so gets them breathing room to improve their own manufacturing process before TSMC can bring on enough capacity to serve both of them fully.
Deeper pockets than Apple?
Deeper pockets than AMD. All other assumptions holding true, Intel doesn't need to outrun the bear...
Please expand.
The pockets aren’t deep enough...

TSMC knows Intel is a competitor which AMD isn’t so any deal they’ll sign will be temporary at best.

They aren’t exactly struggling to find customers as every wafer they have is spoken for 6 times over.

If anything at least as far as TSMC goes Intel would have to pay through the nose for any capacity.

I’m mildly irked by this but only because there will be more than a few people who will believe this to in response to the dreadful letter from Third Point.
If the play is to use a different fab for about 5 years maximum then plough forward with their own, then I think it's a good strategy.

If Intel can get a good design onto 5nm relatively quickly, they can at very least curbstomp AMD's current single-thread lead until their new nodes come online. Intel are something like 20% behind in single-threaded work now, despite being still stuck on lithography somewhere between 4 and 5 times less dense.

Moving away from fabrication is really dangerous, forcing yourself to pull your trousers down and face the music is a good thing in the long run. Ultimately this is a question of whether the leadership have got the bottle.

> despite being still stuck on lithography somewhere between 4 and 5 times less dense

That seems a bit high, I know it's 14nm vs 5nm but they're not real numbers anymore, and even that would only be 3X.

Don't get me wrong though, Intel are still screwed. Even if they get 5nm in 5 years, where will TSMC be then?

You can (and should, I'm slightly surprised it's so uncommon "even on HN") look up the actual transistor density on Wikipedia - Intel are still stuck on 37.5 MTr/mm^2, TSMC cutting edge is much more than that and Intel 10nm is about 100 MTr/mm^2. So the actual number is between 3 and 5 depending on if you compare with AMD on Apple.

You have to realize that all the fabs (and Intel) have multiple nodes on the go at any given time, the progression isn't completely serial - they've basically failed this generation but that doesn't mean that they won't be back up there if they nail the next one.

Intel's 7 nm process has also essentially failed; it's years behind schedule.
I stand corrected! For reference, TSMC's 5nm is 173. I almost feel bad for Intel. Almost.
Why doens't Intel just pay Samsung a lot of money to license their methodology for the latest nodes? This worked for Globalfoundries back in the day.
My biggest fear of fallout from something like this is that Intel could flex their financial muscle with TSMC and squash AMD just as they're pulling ahead. It wouldn't take much for Intel to buy out all of TSMC's available wafer starts...
TSMC has some say in the matter.

If they view Intel as a competitor and AMD as a long-term customer which would seem rational, why would they blindly auction their capacity to the highest bidder of the two?

Frankly I'm more concerned with Apple's deep pockets continuously monopolizing TSMC production.

I think Apple is in some ways a better customer for TSMC than AMD is. Apple has far longer lead times than AMD since they're producing a fully assembled system. AMD seems to be short on supply with the launch of every new generation, and it feels like they have been underestimating demand. It's a good problem to have, but difficult to handle without having piles of cash on hand.
For TSMC? perhaps. For customers? Absolutely not. I have zero interest in a future where Apple produces the only general purpose computers containing state of the art silicon.
I would think TSMC also wants a diverse CPU design market. The higher competition decreases design margins, leaving more profit margin for the comparatively monopolized fabrication.
It would be terrible news if a new CPU/GPU is not in short supply at launch. The short-term short supply generates the long-term short supply :)
I don't think it "wouldn't take much". Intel certainly has resources, but they'd also be competing against Apple, nVidia, and some other heavy hitters. Additionally, I doubt the TSMC would have much incentive to de-prioritize the needs of a valued customer (AMD) against a competitor in the semiconductor space who will likely switch back to their own fabs as soon as they're competitive (if they ever are).
Fabs take time to build. Just look at how long it took Intel to scale up their 14nm capacity when it became clear the 10nm process wasn't going to start working any time soon.

On the other hand, the added revenue would further cement TSMC's position as leader in the process race. Given the resources of so many customers pushing the leading edge, I don't see how Intel could recover anytime soon. Unless TSMC grows to become fat and complacent like Intel did.

I thought I heard Apple already bought out a lot of their capacity?
TSMC is a Taiwanese company, the very public and popular head of AMD is Taiwanese and AMD maintains Taiwanese offices, there's always going to be some bias in this kind of tradeoff that won't get AMD screwed.
It's funny how people never learn. It is absolutely imperative that Intel continue manufacturing. You people are making the exact same arguments American companies used when they outsourced manufacturing to China. Now China is the world's factory and they call the shots as the west kowtows. I don't understand how users here can have such a fear of Facebook or apple while blithely supporting policies that would make you slaves to foreign interests. Y'all are so myopic.