AMD CEO sees chips from TSMC's US plant costing 5%-20% more

If that is true, perhaps the US should stop destroying the western alliance.

The current US administration has been threatening two separate NATO allies with invasion and annexation.

Not even Russia, with their daily Russian last warnings of nuclear Armageddon, dare being that hostile.

Averaging over a large window while ignoring the trend is not reasonable.

Trust is a funny thing like that. You do have to do it all the time, but if you fail even once without extremely good reason you lose it all.

If we look at the military investments US did since Clinton(so, last 30 years), you'll notice a trend of looking after it's own interests before the ones of the world. An example is the lack of investment in destroyers to patrol the seas, while at the same time the focus shifted to super-carriers which are good for one thing: obliterate a single, powerful country.

This is not just Trump, but everyone after Bush Sr.

There is a quite long history of USA doing coups, sabotage, and so on, against its own "friends".

I'm amazed at how many people think China is going to take Taiwan by force. They're playing a long game because they want it intact. They want the people there to want to be part of China. That doesn't seem to be going very well, but how can outsiders know? But again they're playing a long game and have plenty of time so long as things are moving in the right direction.

Planned obsolescence does happen, but the phoebus cartel is the worst 'example' of it.

Assuming you refer to NS2 blowup, it was unused when it got sabotaged.

If you’re going to use accents, technically it’s Táiwān (ㄊㄞˊㄨㄢ)

[1]: Why are these countries “autocracies”? To better resist foreign intervention.

Numbeo shows the cost of living (including rent) is 45% higher in Phoenix than in Hsinchu (where TSMC's 2N is)

Rent is 176% higher.

The part you don't want to be around for is following the realisation that there is no path back.

The main issue here is political.

First, high-end chips have essentially a single global market. Compared to the value of the product, transport cost is negligible. If a TSMC Taiwan factory has an oopsie, all its customers are going to be buying from your local US plant - so you are still ending up having to deal with the effects of a significantly higher demand. AMD unable to ship? Expect Intel to go out of stock rather quickly as well.

Second, the chip manufacturing supply chain. Having a local chip factory is nice and all, but where is that factory getting its supplies and equipment from? Most of it does not come from the US, so during another COVID your local chip factory might still be forced to shut down. This also applies downstream: what use is a fancy high-end CPU if you can't find anyone locally to produce all the trivial parts you need to support it? Who is going to manufacture those trivial-yet-essential $0.05 connectors and $0.001 capacitors or resistors? That has all been outsourced to Asia decades ago.

A single US plant isn't going to do anything for your supply chain robustness. You're going to have to rethink the entire chain and each step is going to be 20% more expensive, so better prepare for a doubling or tripling of the final product price.

Local factories are nice for the defense industry, where the confidentiality needs due to national security might warrant the premium. But regular consumer chips? You'll be paying a huge premium just so a politician can get a couple of favorable headlines, often without there actually being a significant impact to the local economy.

I already own a perfectly adequate computer for my needs. In every possible way this won't affect me, and infact so long as the cheaper product is available for purchase it still won't affect me. If I'm a business I'll be 20% better off then other local businesses by continuing to not buy local anyway. If I'm consumer...well I'll just have more stuff I want.

And so on in this way you might want to go read up on The Tragedy of the Commons in economic theory and then reflect on what one of the primary roles of government actually is.

The most expensive, highest-margin, technically advanced and risky business in the world is for investors who want that in their portfolio. If they wanted to milk a dying industry on the way down they would go buy Disney stock.

It is very clearly in the interests of long-term investors in Intel to maintain a commanding position in fabrication: it's been the secret sauce of the company since the very beginning, it's never been more in demand.

This idea that companies are obligated to do what will deliver some little bump in the stock price in 90-180 days is everything from not how the rules work to just a lazy meme for people who don't want to earn their princely salaries.

Don't make excuses for weakness at the top.

Inflation-adjusted INTC [1] is the same as it was in 1997, including dividends! Shareholders have no real return from INTC for almost 3 decades.

[1] https://totalrealreturns.com/s/INTC

I would quibble with the exact right thing phrases but otherwise agree. Intel indeed followed a formula which is intended to and often does produce massive return for some time frame. The formula is indeed "gutting the company" - squeeze every part of an enterprise and return the results as profits. Whether destroying the companies long term prospects is worth these short term profits is a complex calculation.

A managers' duty is to promote long term value and stability, actually, but return enough short term profits and you trump that long investment income.

In 1998 Meriwether and the rest of LTCM nearly crashed the economy, needed the Fed to get involved, and they were personally ruined, guy never opened a ten thousand dollar bottle of wine again and probably never had anything again. Shortly thereafter, Jeff Skilling took out offices in 9 cities and pension plans all over the country with shady accounting. 24 years in prison (reduced later to 14). Ebbers/Worldcom 2002: died in prison.

By 2008? Zero prosecutions. Bonuses the next year.

Around the same time Clinton got caught lying about chasing (consenting and of age) skirt in the Office: nearly ended his presidency, definitely ended his policy agenda, real consequences and he caught a shooting star to avoid far worse. The public was not going to accept it, Congress was not going to let it slide on either side of the aisle. Today? Something like that barely makes the press. You have to be accused of sex trafficking to even get an investigation started and everyone will probably walk.

The idea that this became uniquely bad in January, or even 2016 is demonstrably untrue. At some time in the last 30 years we started accepting leadership who are dishonest, nakedly self-interested, lie without consequences, enrich themselves via extraction rather than value creation, collude with no oversight, and sell out the public.

This is a completely bipartisan consensus on these norms. Speaking for myself, I think Trump represents a new low, but not by much, he's just the next increment in what history will probably call the Altman Era if his ascent to arbitrary power on zero substance continues on it's current trajectory.

What has much more commonly produced good outcomes in such situations is robust public-private partnerships like the ones that produced the semiconductor industry in the first place. Run the list of innovations in strategically key technology and what will you find at one remove in every instance? The DoD, NASA, the Labs and ATT more broadly, the university system.

It's always a public/private partnership during periods of explosive value creation when the stakes are high, and it's always a private sector capture orgy during periods of extractive stagnation like the present.

The current board is a pack of cargo-culting epitaph writers.

They were wasting a ton of time and effort eagerly trying to convince Apple to put IA into phones despite obvious failures to deliver power-effective chips (Atom being the result of these efforts from what I understand). They were spending a lot of time and money trying to start up like a junk ware app-store thing for PCs that they could use OEM relationships to peddle, as if the PC ecosystem belonged to them the way that Android did to Google or Apple's ecosystem to Apple, not realizing that if anyone has that power it's Microsoft (but they also don't).

It was pretty shocking coming from a hacker/cyberpunk culture where everybody had been dunking on Intel designs for over a decade. (I personally had been waiting for an ARM laptop since around 2000.) A lot of leadership I got to interact with were business/people-people types that truly seemed to believe that the best product boiled down entirely to social perception of status and has zero basis in reality. Basically the company seemed to be high on the Intel Architecture's accidental monopoly over personal computing thanks to PC-WinTel becoming so dominant (and Apple's later capitulation) and seemed to believe that it was all because of their "genius" Intel Inside marketing campaigns (which were pure social status signaling, but with an effect of avoiding price competition with lower-cost IA rivals AMD,Citrix,VIA and holding power over OEMs rather than being responsible for the market situation around IA in the first place).

Maybe something in the Hillsboro/Beaverton area's water? Both they and Nike seem to entirely consist of a diet of their own farts.

There is talk about the next version of Arc using 18a. If it does, I'd expect Intel to move that generation's compute tiles to 18a as well.

Apple farms out its displays to Samsung, a competitor. It's just how business is done.

But so far nothing of the sort has happened for a long time. If feels like ever since Ryzen landed, they have been desperate to catch up but keep tripping on themselves. Losing Apple, while inevitable, has made them look even more irrelevant. They still do decent stuff for the most part but there isnt anything really exciting.

I do like what they are doing with Arc GPUs but it is clear those are loss leaders and it isnt really gaining that much traction.

Alas, this is a story where we will have a better understanding in five years from now.

It's not about how soon 14a and 10a come out, but rather about how good they are when they arrive. 14a will be competing against TSMC A16 in late 2026 and 10a will be competing with TSMC A14 in late 2027. The measure of Intel's success will be whether they are gaining or losing vs TSMC.

On the customer front, I think customers are probably necessary to offset the ever-increasing R&D costs and an extra year or two to work on making their libraries more standardized may be best for everyone.

These days iGPUs run pretty much any game I care to play so it doesn't matter.

Now we are in a different situation. There are several big competitors using ARM instead of x86. The software world is actually transitioning away from x86 in masses. Apple does their own CPUs better than Intel. AMD outsourced production already. Everybody is pumping money into TSMC who are are already ahead of Intel and they are moving faster.

Either Intel gets a really really lucky run with their new technology or they need to split off the foundry business. The government may put it on life support until TSMC themselves may run into serious problems.

The better way into the future may be to split up TSMC in multiple redundant and competing companies.

“If Intel can just get this next node they’ll be sitting pretty” is what people have been saying for over a decade isn’t it?

Just getting the nodes working and producing enough chips has been a huge issue for them, let alone having good chip designs on top of that.

“No one got fired for choosing Intel” has stopped applying. They’re even losing server marketshare, which was their rock.

Intel as a brand may survive in some shape or form but it's not looking good for the foundry.

My most recent computer is AMD Ryzen based, but we just bought an Intel-based Dell for my partner because the price/performance was better than comparable AMD machines at the time, possibly due to a sale. But the Intel chip is a lot faster than my laptop, so now I'm a little bit jealous of the Intel machine.

Nowadays, there will be another process node from TSMC. If AMD doesn’t pay for the R&D, TSMC’s other customers (like Apple and… actually, Intel) will instead.

Part of it, sure, but they were still fabless and in the ditch before Zen. Unless you're referring to going with TSMC instead of GloFo as going fabless.

Intel's fab would be doing much better if it spun it out a while ago and was making Intel, Nvidia, and Apple chips right now.

Yes, they are.

https://www.tomshardware.com/tech-industry/semiconductors/in...

Definitely struggling, but still in the game.

Intel fabs have never had to be as cost effective as others. They were selling top end chips for top dollar for decades. I bet there are 10 other companies that can make 45nm chips cheaper than Intel can on their old equipment. I could be wrong.

How many of them develop high performance x64-64 cores?

Maybe if you ignore they're the only player with remotely competitive discrete GPU IP for graphics and AI, after the Nvidia and AMD duopoly.

I mean for gaming there is already the Ryzan Max+395 which already is beyond the level of low end graphics (at least if placed in a desktop where it's not heat/power throttled). But it's a bit of a unicorn (especially if you look for a system where it can run full throttle).

but I'm not sure about the beat nVidea part, nVidea has some experience with putting ARM CPUs on their graphic cards and as far as I remember on for their server center solutions there is one which pairs up graphic cards (and their RAM) over PCIe and mostly cuts out the CPU

https://www.nytimes.com/1997/05/14/business/suit-by-digital-...

There were much more discussion of Intel stealing things in the 90s than today.

Also, I looked into the claim when he had said it, apparently he was being intentionally misleading about it, and the press tried to ask what he meant: he was speaking tensor performance on future enterprise Arc card products at datacenter scale, ie, AI bait.

In early 2021, Nvidia's compute flagship was the A100, 19.5 FP32 MAD TFLOPs, but the misleading number they quote in marketing is the tensor performance of 312 TFLOPs of FP16 accumulates. That would be about 3.2 million of these at tensor perf.

Skipping H series, in late Nov of last year, their new flagship is the B200. 124 FP32 MAD TFLOPs, 2250 tensor FP16 accumulate TFLOPs. That is now 445k cards to reach zettascale if using tensor cores. You won't be fitting ~1400 GPU-laden machines in a single datacenter, but the number is becoming more manageable.

They improved, in 3.5 years, 7.2x.

Lets say Nvidia does this again. 3.5 years, again, would put you in early 2028, and they manage another 7.2x win: that could be 62k cards across 7.7k. That absolutely is doable in a single datacenter.

The problem is, and this is where the prediction actually falls apart, not that its impossible: We don't know what future Arc cards look like, nor enterprise ones. Battlemage is an improvement over Alchemist, so the tech *is moving forwards at, but either Celestial or Druid was supposed to introduce the enterprise compute card variants, but that seems to be dead, and no indication either of those lines will even see the light of day now. The new CEO seems to be hard set on making Xe for iGPU only.

I can't find any hard numbers on Intel's tensor units, but apparently they're actually competitive. I can find the normal FP32 MAD numbers, and it ends up that Intel is 13.5w per TFLOP and Nvidia is 8 and both companies have equal efficiency in transistor usage. Assuming Intel made a B200 competitor, and assuming the higher power usage is due to voltage (Intel B series voltage is similar to Series 40's voltages, which is a lot higher than equivalent enterprise/pro series cards), Intel could be making a card that's somewhere in the ballpark as 2/3 as good for the same power usage.

So, in the end, yes, I don't agree with his claims of future Zettascale at Intel by 2027. I don't think he was wrong for the industry as a whole, however. If he would have said, say, 2030, I don't think we would be discussing this, that certainly would have been doable if he was at the helm and they kept doubling down on Arc every gen and everything went according to plan.

Betting some on Intel is very wise when the alternatives are, as I see it: 1) investing in TSMC building fabs and creating more of an employee knowledge base and skill base on shore, 2) hoping a US-based startup gets enough traction to grow.

Agreed on leadership. But selecting leadership teams, especially technically-grounded leadership teams is extremely difficult. Which is why companies revert to non-technical leadership so often.

Since Intel has been mismanaged for so long I don’t know how many good lower level employees they managed to retain, I doubt much would be left if they properly cleaned house.

TSMC is making fabs in the US, but they are not SOTA fabs. Those are kept in Taiwan.

That being said the government will likely not allow them to fail completely out of the foundry business for geopolitical reasons

There wasn't any bailout on them, what do you mean?

That is the political calculation, not "throw good money after bad" kind of economics 101.

But I guess "too much socialism"

https://www.computerenhance.com/p/an-interview-with-zen-chie...

No, it's not. In modern high-speed CPUs, many instructions are decoded directly, without going through the microcode engine. In fact, on several modern Intel CPUs, only one of the instruction decoders can run microcode ("complex") instructions, while all the other decoders can only run non-microcode ("simple") instructions.

It would be more precise to say that it's at the "front-end" part of the core (where the decoders are) that the ISA lives, but even that's not quite true; many ISAs have peculiarities which affect beyond that, like flags on x86.

US has become less dependable but a lot of countries still depend on US maybe less but not as allergic as depending on China.

Hyundai makes cars and military weapons and probably thousands of other things that aren’t even related to each other, don’t know if they still make computers.

I haven't been to Asia in a while, but at one time, Hyundai made both computer chips and bulldozers.

Mitsubishi once made computer chips, and had a bank, and an art museum.

There are companies that own both department stores and subway systems.

America used to have a fair amount of this, but it was more common during the Industrial Revolution. Companies that owned both railroads and summer resorts. Oil wells and banks.

Even as recently as the 1990's there were companies that owned both pipelines and fiber optic networks. Toasters and television networks.

What this administration is doing is not a recipe for success: trade wars with everyone, immigration crackdowns, and unpredictable tariff policy.

EDIT: Oh and hinting at invasion (Greenland, Canada) doesn't help either

This is all assuming you can get past the NIMBYs to build the plants in the first place.

CPU space is definitely easier to disrupt but the GPU space requires a HUGE investment and you're fighting uphill against proprietary technology like CUDA that has become industry standards. Intel, Qualcomm, Samsung and Google have made inroads with budget to mid range which is the highest selling segment. But to compete with Nvidia or AMD on the high end you either need a whole datacenter or many years of R&D with very little return for a long time. Apple would be on this list but they have siloed off themselves entirely.

What’s positive is that we have state of the art domestic manufacturing with potential to onshore more and more of the required supply chains, building/educating local expertise, etc etc.

It’s silly to focus on shipping.

[1] https://www.tomshardware.com/pc-components/gpus/spitballing-...

In Europe or Japan I can already build much cheaper, workforce is cheaper, now even hardware is cheaper.

How does data centers in US stay competitive? Why would Google or others build their new AI infrastructure in Ohio rather than elsewhere?

For the sake of argument, if all goods increase in price by 20%, Americans are going to have the experience of being worse off than they were before.

This is the largest tax hike on Americans in modern times (possibly ever?). While it may take a while for people to understand the impact of policy, people generally do not like large tax hikes. I don't think it's a stretch to think people will not like this tax increase, either.

This is a confluence of the previous administration having the forethought to do this, before the current administration tried to kill the CHIPS act.

If they hadn’t done that, you likely wouldn’t have seen domestic production able to satisfy the needs.

Tariffs alone are a misguided cudgel.

Also your comment about a “major impact to the consumer” ignores that this is an increase in cost just for the silicon. There’s a lot of tariffs on different parts of the actual product.

If you want to bring back manufacturing you need to consider the entire supply chain, and make sure that inputs for whatever factory you are bringing in are secure and will be equal or cheaper in 20 years. These things need to be predictable.

Also, let's not pretend the MAGA movement listens to anything other than the propaganda.

When you offshore a production facility, it's not about being profitable, it's that it can be more profitable elsewhere.

If you have no guarantee that the tariffs will still be there to artificially maintain your profitability so high, then you don't build.

I think we can agree on the facts there