Intel Reported To Be Looking At Acquiring GlobalFoundries

[mangeek]

I wonder if they're going to get federal dollars (or tax breaks) to 'onshore' manufacturing when they're really just acquiring existing assets here with no net difference to the domestic economy.

[edwaleni]

- AMD's contractual commitment for volume through GF has been fulfilled. AMD can use whom they wish
- GF has already sold off the wafer/design operation to On Technology, so they are already a straight fab operation
- GF has margins, but due to competition, just enough to cover their cost of capital
- GF would need a very large capital infusion to ramp a new node, it's why they dropped 7nm is they couldn't get enough volume to recover the capital costs

The entire fabrication industry is currently going through a process of dis-integration. Intel is incredibly vertically integrated and has been for some time.

But the issues around supply chain and sourcing along with this movement to dis-integration is forcing their hand to move tactically via acquisitions.

So Intel buys GF, turns them into a "fab-on-demand" operation along with supporting SiFive. It allows them to move with the market while they sort out their own internal vertical integration issues.

[Vistaus]

And what about buying AMD, so they finally have decent CPUs and GPUs ... and not that trash they are
selling since lots of years ... (needing mitigations which eats a good deal of performance, ME/CSME, ...).

Sure, that would've solved the problem because AMD didn't make such mistakes… oh wait.

Don't get me wrong: Intel pulled much more crap, but AMD was also affected by some vulnerabilities, so it's not like buying AMD would solve all of their problems.

[JMB9]

You're talking about a monopoly. Sure, go for it if you only want one player with prices higher than what they are now and low quality.

Sorry, but Intel destroyed itself - no technical person would buy an Intel CPU - it is still the same crap which gives data for free - while others (like AMD - but not limited to it) have problems with meta data. I don't think that any technical person with focus on Linux would buy an Intel system since several years.

I had hoped that Intel would come back soon - but still only crap for the desktop (and similar for servers ... security is there even more important than on the desktop). And yes - I have experienced what mitigations made with Haswell - so this is no way to the future.
And now - instead of really making a reasonable fresh CPU design to be competitive again - they buy other companies ... very reasonable. I don't think that Intel got a sane management - it seems the same course as before (titanic anyone?). If this would be a technical world, Intel could not even afford to buy anyone - but as shown by apple it does not matter at all.

So Intel does not exist for most Linux users - AMD has something of a de facto monopoly on Linux right now, as IBM and ARM are both not ready (yet) to be competitive. I hope for competition - but Intel is just not willing to play the game - so they are out.

[JMB9]

Sure, that would've solved the problem because AMD didn't make such mistakes… oh wait.

Don't get me wrong: Intel pulled much more crap, but AMD was also affected by some vulnerabilities, so it's not like buying AMD would solve all of their problems.

Are you serious - no one would compare the big data leaks present in Intel CPUs to the problems AMD, IBM, ARM and others have.
Intel just lit the candle on two sides - and burnt their customers. And they are using this crap design till today - and use a plethora of mitigations which is a security nightmare.

This can not be compared in any way to other CPU vendors - no one comes close.

[Ironmask]

Sure, that would've solved the problem because AMD didn't make such mistakes… oh wait.

Don't get me wrong: Intel pulled much more crap, but AMD was also affected by some vulnerabilities, so it's not like buying AMD would solve all of their problems.

It still speaks volumes when AMD, the clone manufacturer, made their chips more secure and patchable than the very company that originally created the architecture.
(And don't mention how AMD made AMD64, it's still a backwards-compatible architecture created from x86)

[piotrj3]

"
Brane215
Nm claims below 28 nm are pure martketing bulls**t.
"


you both are wrong but it only shows you are not business insider you do not unterstand what these numbers mean.
yes 34nm SOI (of AMD bulldozer chips) and 45nm (for intel) was the last 2D chips
if you watch the most advanced node on this planetr earth the IBM 2nm node then yes you can not find on a microscope any number like 2nm or 5nm or 7nm.... because if you calculate in 2D the structures are much larger...
this means IBM 2nm node in 2D speech is more like 7-10nm node...

but no one cares about 2D structures anymore and this is the point you are and also brane215 is wrong.
per 1mm² the IBM 2nm node has 3 times higher count of transistors than 5/7nm and every 2D transistor is in fact 3 transistors in 3D....
any business insider know that below 34/45nm you have to calculate it in 3D and not in 2D anymore.

and if you do calculate it in 3D then yes IBM 2nm node is in fact 2nm in 2D if you compare it to 34nm SOI-.--

the 2nm IBM node beats the 34nm in any meaning in power consumsion per performance in the ability of higher frequencies in the ability in lower power consumsion in idle and also in the max performance.

so really no one cares about people Brane215 and piotrj3 people who think 2nm 3D node is not better than 34nm 2D node.

people like this have no fucking clue at all. 2nm IBM node beats everything from 5nm TSMC to 10nm Intel...



last time i checked it intel 10nm is not in any meaning similar to 7nm TSMC....

but even if this would be the case TSMC has stopped to produce in 7nm did you not know it?
today TSMC node of choice in this facturies is the 6nm node. what has 28% more transistors per 1mm²
and yes again in 2D the 6nm is not better than 7nm node but in 3D you get 28% more
no one cares about 2D nm numbers anymore.



no intel does not fine. AMD use 6nm and 5nm now and intel has no fucking clue what they are doing to compete against this or even better IBM 2nm Node... Intel has no fucking clue how to compete against this.

and remember TSMC 7nm and 5nm if you read on wikipedia was orginally developed by IBM
this means IBM develop the 2nm node and then they licence it to TSMC...

IBM/TSMC they have 2nm/3nm/5nm/7nm right now and intel can not even produce high performance chips (mhz) on 10nm they can only produce low performance (low mhz) chips on 10nm... they do desktop on 14nm because they need 5,3ghz to compete.

I want to facepalm.

First : YES IT IS MARKETING TERM.

SECOND: change your damn sources.

Third : you are even more wrong because TSMC has 3 diffrent 7nm nodes, HPC, "mobile", and EUV. And all of them have totally diffrent densities.

My source: https://www.techcenturion.com/7nm-10nm-14nm-fabrication

Actually Intel 10nm have higher density then both HPC and "mobile" TSMC 7nm nodes. Only EUV+ is slighty better. And main reason why AMD became so much better from Zen 2 to Zen 3 is actually that diffrence between HPC and EUV+ (which is huge) despite them being both "7nm".

And AMD doesn't use 6nm, there is no one in world planning it. And Intel also bought capacity at TSMC for 5nm.

And lastly, main reason why AMD is winning is chiplet design. Intel has very big problem scalling their desktop chips beyond 8 cores and server chips beyond 40 cores. Meanwhile AMD can simply bring many 8 cores chips glue them together with infinity fabric and producing those 6-8 core chips is actually easy. Intel because they are behind in number of cores have to respond with increased clocks, but increasing clocks is not good for price/performance.

And 2nd biggest reason why Intel is losing in some fields is not providing 10nm chips for desktop users. 10 nm of Intel only exist in mobile market (there they compete very well with newest CPUs both in multicore and single threaded), and server market (where 40 core best Xeon can have extremly impressive ML performance, but still loses in most tasks simply because it competes vs 64 core epyc).

[Developer12]

WTF are you on about ? GloFo is not bottom player. They just aren't going to the cutting edge. Their 12/14 processes are very decent and they are making a fair buck on htier segments.

GloFo has a reputation for being bottom-barrel when it comes to making logic. I got first hand experience of this when working for a while at one of *those* two top-level tooling companies.

Edit, see [1] for a short history of the money pit GloFo has become for IBM, with their repeated failures even given IBM technology directly.

As a senior engineer who had been there many years put it to me: with Intel you could expect a transistor to have a nice square shape with flat vertical sides. On the same size with TSMC, you'd get more of a trapezoid. That's largely why TSMC 7nm is roughly equivalent to Intel 10 in terms of feature quality.

Then at the bottom you'd get GloFo and Samsung's (logic, not memory) process(es), which were about equally crap and would basically just make a lump. The only way to go worse than that was to go to some shitty cheap mainland fab.

Compare GloFo's 14nm to Intel 14nm. There's absolutely no competition. It's laughable.

Note that with Samsung developing reasonably decent 8nm and IBM gifting a really nice sub-3nm process to Samsung for future POWER parts, this is set to change a bit.

[1] https://www.talospace.com/2021/07/in...ower-chip.html

[Neraxa]

I have no idea what they are up to. GlobalFoundaries gave up on small nodes. So you will end up combining 2 foundries neither competent with small nodes? Or maybe if they think if they combine two struggling foundries that it will make a strong foundry? Maybe they finally know what they are doing, or maybe not.

[ddriver]

It does if you've got the money to do it - and Intel does.

The important part is that they DON'T have the money to buy TSMC (and to buy off all the politicians needed globally to let a deal like that go through), so it basically "doesn't matter" for at least another 4-5 years anyway.

What they can AND are already doing is paying off tsmc capacity to limit amd's ability to take market share. And there intel was just a couple of years or so, mocking amd for having to rely on 3rd party foundries.

Amd was essentially tricked out of its fabs and it wasn't the best deal for them, but they did manage to make the best out of that situation, amd's business and margins are higher than ever, and they are beating intel even in the last couple of niches where it still had advantage.

To be honest, amd would have likely gone bankrupt trying to move their fabs from 14nm, it is expensive and tricky, and it makes all the difference ROI-wise what your total manufacturing volume is. Amd's fabs just didn't make enough chips to invest in cutting edge process development. Even intel is struggling for half a decade to move away from 14nm.