AMD Expected to Become TSMC's Second Largest Customer | Hacker News Reader

AMD Expected to Become TSMC's Second Largest Customer

tomshardware.com

153 pointsrbanffy105 comments

If TSMCs march continues, how long will it be before Intel is forced to use them for all their performance chips? Will everyone just have to switch to TSMC?

And if that happens, how long will it be, with much less competition, for TSMC to get sloppy like Intel did, and everything stagnate again.

> Will everyone just have to switch to TSMC?

Did you read the article ? AMD is going up as a % because a lot of customers are switching to Samsung (and Huawei sanctions)

The fact that AMD had the best-performing CPU by a 2x factor for the past 4 years probably plays a role too.

That seems at least a bit hyperbolic - AMD has pushed intel a lot in the past four years but it wasn't until the past year or so that they have potentially started outperforming Intel (depending on workload) in what is likely to be a 'who has released more recently' kind of competition for a while.

https://www.tomshardware.com/reviews/cpu-hierarchy,4312.html

Latest benches have AMD and Intel winning in about the same number of tests.

The first graph in this article doesn't include Ryzen 5000 series processors.

By their own tests later AMD has top 4 positions for both single-threaded and multi-threaded workloads. Intel came out with their Rocket lake line of CPUs after Ryzen 5000 and Rocket lake can't match up.

Even before Ryzen 5000, anything that wasn't single-threaded Intel could never beat; and had way worse performance per watt.

Weird, it definitely includes them for me, in all six images and it's exactly like the parent said (some Intel, some AMD).

Maybe you are confused because Threadripper is at the top in some above the 5000 series?

AMD is killing it with the 5000 series, but it's just barely edging out Intel in these benchmarks. The whole 2x figure was a little absurd.

It’s been true before that the losing architecture can juice their numbers by making big hot chips that use tons of power. Intel is so far behind that this is strategy is hard to continue. They can keep in the conversation by giving you more cores for less dollars, but those are dire straits for the company.

Its hard to continue but they have just posted great financials so only need to scrape through to 10 or 7 nm then they will probably go back to stomping AMD. AMD's documentation is poor, software is poor, marketing is also not great (My next computer will be an AMD but I wouldn't bet on them having a run anything like Intel's over the last decade)

AMD most definitely started outperforming Intel in performance per dollar in the last year forcing Intel to implement huge price cuts to remain competitive.

The original Zen architecture released in 2017 was not the best performing CPU on the market by any metric. It wasn't until Zen 2 and Zen 3 that AMD fully caught up.

If $/perf on multithreaded workloads is your metric then it did beat Intel from what I remember.

I believe it took until Zen 2 for the AMD high end to beat the Intel high end in multithreaded perf, and until Zen 3 for AMD to take the single threaded perf crown across the board too.

It occurs to me now that AMD probably would have taken the multithreaded crown from Intel with Zen 1 if Intel hadn't upped their (consumer socket) high end from 4 to 6 cores in the same year.

Although it's hard to say whether that was a direct response to the threat from AMD or not.

Doesn't that correspond to the globalfoundries contracts?

I’m out of the loop. How does Huawei (PRC) sanctions end up negatively affecting TSMC (ROC) sales?

TSMC's second largest customer _was_ Huawei. The US sanctions led to TSMC no longer selling to Huawaei.

I see, that makes sense. Thanks!

What a lost opportunity though. Taiwan and USA could have made an exception here on some made up grounds and then back-doored the Huawei silicon.

(Not saying they should do this, but that it is something they would do and it’s odd that they didn’t.)

Considering the whole point was somewhat shortsighted pain to China (in that long term this forces China to build a world class chip foundry now), it makes perfect sense. Huawei (and by extension China) is not going to be able to produce as many chips as competitively as others.

>in that long term this forces China to build a world class chip foundry now

Doesn't make any difference, China was planning to do that all along years before, with or without the sanction.

Planning to do it soonish, vs ‘must do it immediately and a 100 billion dollar industry is on the line’ are pretty different looking endeavors IMO. It moves up timelines and forces reprioritization of funding and other resources on a massive scale.

Still might not work though, and with this ban it makes the stakes much higher for China if it doesn’t.

China were planning wasn't the right word. It should be "they were already dong it" . They burned through 100 Billion ( USD ), NAND and DRAM are coming up soon, although whether they are competitive remains to be seen. When DRAM and NAND were extremely pricy they could afford to have low yield, low quality NAND and DRAM and sell it at break even, now the market is back to "normal" they are back to where they were. All of that was from 2013 - 2018. Way before Western Media even had their radar on the industry.

No the point was to protect national security interests by making sure that Huawei devices were not used in our infrastructure so they couldn’t back door us. It was implemented with a heavy-handed “don’t do business of any kind with Huawei” sanctions regime. In retrospect, supplying to Huawei should have been allowed.

So why pressure other countries to drop Huawei too? The UK and EU countries previously thought Huawei's presence in their 5G infrastructure was a manageable security risk, until the US pressured them. By all means, it appears to be a "starve the beast" campaign to delay or deny China technological primacy in 5th-gen telecoms, and hinder the "China 2025" agenda in general.

Among other reasons, it’s a bad idea to let allies, with whom you share intelligence, build their network layer using a strategic rival’s hardware.

Hardware back doors are basically impossible to find and remove after the fact.

> Among other reasons, it’s a bad idea to let allies, with whom you share intelligence

That's a non-sequitur: classified information is not transferred unencrypted over public networks. Isn't it a little conceited to think other countries are unable to make their own intelligence assessments, when presented with facts? The reasonable conclusion is that the considerations went beyond just security - because the UK in particular had published a public version of a report on 5G security vis Huawei. Ultimately, the decision to drop Huawei was top-down,from politicians, after a lot of diplomatic pressure from the US, and not from career intelligence people, some whose day jobs I assume includes threat assessments.

You’re right that nobody (hopefully) is transmitting sensitive information in the clear, but having a network layer that you don’t trust fundamentally changes your security posture for the worse.

For example, consider the possibility that your 5G equipment might go down at an inconvenient time if there’s an actual military conflict.

Also, it’s not always clear what information has strategic value to your rivals. Even simple metadata, like which political leaders are talking to whom, can be revealing if collected at scale.

Huawei end up having to use SMIC, so less business for TSMC.

SMIC barelly can do 28nm chips, so Huawei is actually out of consumer mobile business for quite a while. No one that counts will produce chips for them, no one will sell the technology. With SMIC they gonna have to expand a line of products demanding less computation power in order to survive.

28nm is perfectly fine for networking gear, which is arguably still a bigger business for Huawei.

>how long will it be before Intel is forced to use them for all their performance chips?

One year? https://www.trendforce.com/presscenter/news/20210113-10651.h...

New Intel CEO Pat Gelsinger To Host Webcast [1] About Intel’s Future On March 23rd. We will see if we have more details then. But Pat seems to be committed to the IDM model.

>for TSMC to get sloppy like Intel did, and everything stagnate again.

I dont see that happening any time soon. Not in this decade. Not with the current and upcoming management of TSMC. Even assuming the worst of a culture change within TSMC it will still take time to rotten to its core and time for it to manifest. Intel loses out has nothing to do with technology, but the combination of pathetic management and record making profits at the same time for years.

[1] https://www.anandtech.com/show/16560/intel-ceo-pat-gelsinger...

TSMC will probably have a second largest customer even if their process development stagnates.

The sloppyness isn't an issue. Lack of competition that can keep up is. Especially in a geopolitically charged environment.

TSMC seems like the number one security guarantee for Taiwan.

How does this change with TSMC opening a fab in arizona? (And Samsung in Austin)

The TSMC fab in Arizona as announced is 20k wafers/month, their total production is over a million a month. It's a tiny drop in the ocean.

there is this announcement: https://www.bizjournals.com/phoenix/news/2021/03/02/taiwan-s...

Still relatively small, but closer to an order of magnitude than two.

I believe this branch of TSMC is to ensure that the American supply chain for weaponry remains intact even if something happens to Taiwan.

edit: Apparently the plant is expected to be a 5nm fab

Well this is pure speculation, but that branch of TSMC is going to be an American company. As it will definitely be totally dependent on its mother corporation, making problems for that mother corporation could make problems for the American company - so IMHO the USA will protect the mother corporation more than they did up until now.

The Arizona play is most certainly a concession demanded by the US gov’t. The US has a major problem with dependence on chips made elsewhere - if China started shelling Taiwan and put an embargo in place for electronics, the US would have major major issues in very short order. It isn’t JUST Chinese chips of course, but that is a pretty visible one for sure.

The entire semi industry depends on Taiwan.

No fab in the world these days works without some single vendor on the planet consumable from Taiwan.

So, the entirety of semi industry will likely go down as it is today.

that's my speculation too, but amended with the provision that "it will be Nationalized iff taiwan gets invaded" (kind of like what happened to Merck, during WWII).

How many American soldiers station in Taiwan? And more importantly: how many nukes?

Since the answer to both is "zero" TSMC actually decreases Taiwan's security - it gives China a strong incentive to attack, with zero means to prevent it on US side.

There are, and have been, US troops unofficially stationed in Taiwan the whole time even after the official bases were closed in the 70s. The US recently expanded the footprint of AIT, their technically-not-an-embassy facility in Taipei and built another in Kaohsiung. US personnel are present in some capacity at basically all ROC air force bases and at the over-the-horizon radar facilities used to monitor east Asia.

This has been an open secret for decades, recently causing a bit of a laugh when a US serviceman was in the background of a publicity photo of president Tsai Ing-wen touring a radar installation.

US policy of not deliberately provoking China over the Taiwan issue doesn't mean the US hasn't maintained a policy of nuclear deterrence over Taiwan since the Eisenhower administration. The PRC and US are well aware, and have extensively war-gamed, that a US/China military confrontation will very likely escalate to nuclear war. The US sailing a carrier or cruiser through the strait every time the PRC makes loud noises is reaffirming the status quo.

My (non-expert) view on this:

If China attacks Taiwan (and the TSMC fabs are destroyed), everyone loses, including China. In the short term, it would probably hurt the West more, but once fabs in the West start spinning up, China would risk being embargoed from the most efficient chips, putting them far behind until they can catch up. (All the lithography machines seem to be coming from ASML in the Netherlands).

This may change in 5 years or so when China has caught up more, and is less dependent on TSMC chips. And China has patience. They'll deal with Hong Kong and Xinjiang first, ensure independence, then invade.

But what if they attack, or otherwise more subtly take control of Taiwan, and the fabs are not destroyed. I'm sure that would be their preference.

Demolition charges are said to be in place to ensure this doesn't happen.

>it gives China a strong incentive to attack

That can be mitigated by planting bombs at the TMSC factory. If china attacks, they'll get end up with a rubble.

This mitigates only half of the problem - yes, China won't be able to use the factory for themselves, but if you take a closer look at the list of the largest TSMC customers from the article those are mostly American companies. Destroying TSMC factory would hurt the US disproportionately more than it would China. Which might matter in the current China-US trade war.

Also "planting bombs at the factory" scenario is one of the reasons why all armies in the world have spies and special forces - you send your special ops soldiers before your main attack to take control of the facility and prevent initiating the self-destruction protocol. I'm not saying it always works, but China can take a calculated risk here, since they win either way.

And what would China win exactly? Even if they somehow miraculously manage to keep TSMC up and running they would not be able to export anything produced there to other countries and permanently loose access to all western IP.

And the sanctions from US and other countries would massively outweigh any perceived economic benefits from invading Taiwan.

Any scenario where China invades Taiwan, global sanctions and embargoes are already happening. Blowing up the wests high end chip fab capacity for a minimum of 2-3 years at that point is a bonus.

If China attacks Taiwan, those American customers are going to lose TSMC as a supplier either way.

That would freeze the car and smartphone production worldwide. No country would like to see this.

And why would taiwan care about worldwide car/smartphone production when the PLA is marching towards their factory?

Taiwan won’t necessarily care, but intelligence agencies of every other country undoubtably would. And putting proactive mines in a facility like that would be noticed.

Rubble, yes, but people that can rebuild it also.

But most importantly, it would paralyze everyone except them economically, as they are the only country that can't use TSMC.

Why wouldn’t those people just move to different countries? I’m sure that most countries in the west would be very happy to receive large number of highly qualified engineers for free.

China can’t really close them in labour camps and expect them to be very productive...

I’m sure it would paralyze everyone economically, since China would just end up with a bunch of ruined factories with no staff and no way to replace most of the machines needed for fabrication (which are made in other countries).

simple replicating everything TSMC is doing in China would be several magnitudes cheaper (if you factor in the economic outfall of a military invasion)

I’m sure they are already duplicating (or attempting to) everything in China. They’d need to stop new research and development though to not always be a few (or more) steps behind. Blowing up the fab and putting all the engineers in a prison camp would do quite well at accomplishing that, no?

You don't need them to be productive. You just need them to give you some secrets, and then your engineers continue the rest.

> it would paralyze everyone except them economically

It would cause a lot of economic damage to PRC too.

If they attack Taiwan, definitely the US and its allies will retaliate. They'd likely place trade sanctions on PRC which would damage much of it export businesses. They might even consider a naval blockade of PRC.

A naval blockade of a nuclear power probably leads to a nuclear war.

Taking Taiwan is a giant pain in the ass for China which is why it is unlikely to happen in any foreseeable future (e.g. 30-50 years). Honestly, it seems like China despite its words sometimes is perfectly content with the current status quo so long as Taiwan doesn't try to foment political unrest on the mainland.

> A naval blockade of a nuclear power probably leads to a nuclear war.

Suppose PRC tried to blockade Taiwan, US might respond by trying to blockade PRC. In that scenario, what does PRC do? Nuke US? Try to work around the US blockade? Come to negotiating table?

I don't think "Nuke US" would be at the top of list of strategies for PRC. Nuke US, US nukes you back, game over. Not a winning strategy.

PRC is probably the one, and only state which may survive a full scale nuclear exchange.

Americans not taking it seriously is reflecting on the overall direction, and quality of their military decision making.

The PRC could survive a nuclear exchange with anyone except the USA or Russia, as both went completely overboard during the cold war arms race. Both superpowers have been constantly disarming since the 90s and there's still an order of magnitude difference in number of nukes between them and the rest of the world.

It can.

And it would've been true even with cold war peak weapon stockpiles.

To anybody claiming the opposite, am asking them just why wouldn't it be any different how in Mao's most famous saying.

Even with an extreme prospect of loosing half of population, you will still have a very angry country that too went completely overboard on preparing for NBC war, and which will still have world's biggest heavy industry even after massive losses.

Even to this day, NBC shelters are built in every Chinese city. Xi, and top military officials spends most of their day in bunkers. The industry is holding bi-monthly, or tri-monthly mobilisation, and war production drills. Most state employees get military training, and basic military training is still a part of high school, and college in China. Humongous food, and materials caches across the country prepared explicitly with war in mind.

And of course, the famed Kim style VIP nuke shelters in the mountains to hide a big portion of the party elite are very real too, as are the draconian plans which CPC has for its population in case of war, with being shot on sight for opposing conscription probably being the least cruel.

Seeing well educated, if not elite, people in the West dismissing that so easily makes me nervous at least.

The current plan seems to be a takeover from within, using Chinese sponsored agents and social movements.

It probably won’t let them march troops on the island for a long time - if ever - but I bet they’re getting plenty of intelligence and trade secrets from it.

That's what people said about Hong Kong but things shifted quite fast.

AMD's success is great for the industry.

x86 and GPU competition is much needed with complete dominance from Intel and Nvidia.

Why we can't build chip factories in the west ? It seems we are bound to have our capacity limited by a few factories that can build all electronics we need. Nobody see the lack of availability of chips around ? I guess Asia is indeed the new axis of world power.

We have lots and lots of chip factories in the West. A very important one just lost thousands and thousands of wafers because the Texas power grid went tits up, and it set the companies using it back by months. There's foundries being built all over the world. TSMC just happens to be the first to have an extremely expensive, extremely difficult process down to a reliable production setup, so all the orders that absolutely have to be on that process are going to them. They also happen to employ most of the world's expertise in leading-edge lithography tech. But if you look beyond the absolute pinnacle of production processes, and into the >100nm range, there are literally hundreds of foundries around the world, many of them in the US and Germany for example, churning out what was top-of-the-line a decade ago.

US and Germany specifically even have what was leading edge just a few years ago. AMD's Zen 1 was made at a GloFo fab in New York.

Asia has been the primary axis for semiconductors since the mid-1990s when Silicon Valley ceased to be about any silicon. That's when the entire industrial ebbed out of the USA.

Were you not paying attention as all the jobs were lost?? Did you not notice the changes?

It used to be that 90% of all semiconductor companies were based in Silicon Valley, hence the same. Driving around the Valley was a near-religious experience as you saw EVERY company name you ever heard of and there was their building.

That no longer exists. And even social media companies are remotely as plentiful today. There's no comparison.

The only hope now is:

* TSMC is building a fab in Phoenix * Samsung is building a fab in one of three areas: Phoenix, Austin or Malta - they haven't decided yet * GF is building another fab in Malta

NONE of this will EVER be in California ever again let alone Silicon Valley, however. It's no longer economically viable there.

Yep. When I was growing up I lived across the street from a superfund site in San Jose. And there's still an active superfund site in the middle of Palo Alto.

Samsung already said Austin, not far from their other fab in Austin: https://www.theverge.com/2021/3/3/22311825/samsung-texas-aus...

> Why we can't build chip factories in the west ?

We have tons of them. TSMC is just extremely prominent right now because Intel stumbled on their 10nm upgrade and is stuck a generation behind.

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

Intel foundries are also less prominent because they don't sell capacity to others... AMD, Apple, Samsung, etc., would never have their processors created at an Intel fab. TSMC is neutral.

GlobalFoundries (formerly AMD) balked at the expense of getting 7nm fabrication going only to have to compete (on price) with TSMC and decided to stay at 12 nm (and larger) for a couple years. But sooner or later they will have to move forward, and perhaps find themselves ahead of TSMC for a while.

Semiconductor manufacturing is pretty dirty. Lots of toxic chemicals.

https://www.bloomberg.com/news/features/2017-06-15/american-...

We can and do make chip fabs in the US. Intel choose a different process for its 10nm than TSMC 7nm. It hasn’t worked and instead of switching to a similar process the other foundries use they doubled down repeatedly. It’s the perfect example of sunk cost fallacy.

>We can and do make chip fabs in the US.

But just barely, and not state of the art...

There is one planned to be built in US https://www.anandtech.com/show/15803/tsmc-build-5nm-fab-in-a...

Also GloFo has fabs "in the west" https://en.wikipedia.org/wiki/GlobalFoundries#Fabrication_fa...

GloFo has more or less given up on cutting edge node design, if I recall.

Fortunately EU is finally putting money on the table :

"EU Signs €145B Declaration to Develop Next Gen Processors and 2nm Technology " https://news.ycombinator.com/item?id=25602950

This is tax payer money that people had to forcibly part with to use to fund private businesses that will not share their profits with those people. Is that correct? Why people cannot directly buy shares in European semiconductor companies and then profit like other shareholders do, but it has to go through bureaucratic channels (and bureaucrats also take their cut for facilitating this bureaucracy)?

> Why people cannot directly buy shares in European semiconductor companies

Because this way the companies can be directed towards a specific desired outcome that benefits society instead of the shareholders.

Sounds like it will benefit shareholders just fine?

In the end they always win.

Even if they need to be dragged there kicking and screaming.

Pretty sure Enron shareholders were left holding the bag?

Historically, Silicon Valley was created from government contracts and subsidies.

Because the European innovation ecosystem is terribly broken and our politicians think regulation is the solution to all problems.

The problem are high taxes (for employees) which means it is expensive to retain talent plus mountains of bureaucracy and more taxes.

You’re saying that China doesn’t have bureaucracy or taxes? Their income tax seems to go up to 45% so it seems unlikely that nobody else in the world can match it. South Korea is 42%, so how does Samsung manage?

In China at least, you make sure you have the right party members on your board and all that pesky paperwork and taxes mostly disappears. Not sure about Korea.

Workers in China can't just pack their bags and leave to a country that values them, so for them it is what it is.

How many Americans give up their citizenship to avoid paying taxes? It happens but it’s extremely rare and I’m highly skeptical that tax rates are the explanation here.

West was or even is sleeping on Asia

Ask average IT person about big IT companies and nobody says anything about Asia, which is scary as fuck.

Missing the word "leading Edge" in the first sentence.

And it wasn't long ago US had GF and Intel.

We can, TSMC opening a fab in arizona and Samsung opening a fab in Austin.

75% of Intel's production in US based, no?

Some. Not that much as in the 1980s. Notice that most of the floorspace is Arizona, Oregon or overseas.

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

Back in the day I worked at Fab 2 but that's ancient history.

Interesting that their current 3DXPoint/3DNand fab is in China and a 10nm fab is in Israel.

The US just gives away secrets now?

Intel is a private, global company, not a US secret.

And SMIC is just another competitor. Right?

Huh, I'm surprised this wasn't the case already. Apparently HiSilicon was the second largest before.

Guess I underestimated how much mobile chips there are, and overestimated how many Zen 2/3 chiplets + GPUs + consoles there are.

Apple and Huawei were typically TSMC’s biggest customers and both those companies have been on the bleeding edge driving process technology (typically adopting the new node 1 year before everyone else). With the Huawei ban, it’s unclear if their absence as a TSMC customer will affect 3nm.

I don’t see AMD having a broad enough portfolio or deep enough pockets to chase the leading edge nodes in the same as Apple and Huawei.

Interesting fight between TSMC and Samsung as foundry of choice for semiconductors.

Global Foundries have skipped a step.

TLDR: AMD is becoming a bigger customer to TSMC because others are rightly hedging their bets with manufacturing at Samsung.

AMD is (rumored to be) as well https://www.gizmochina.com/2021/02/02/amd-outsource-gpu-apu-...

Reply on news.ycombinator.com