Hacker News Reader

Comment by Aromasin | Hacker News Reader

Already being planned by Intel. They're building a new leading edge fab in Magdeburg, Germany [1]. This is backed by the new European Chips Act [2]. Wheels have started turning very quickly with regards to home-grown semiconductors in 2022.

[1] https://www.intel.co.uk/content/www/uk/en/newsroom/news/eu-n... [2] https://ec.europa.eu/commission/presscorner/detail/en/ip_22_...

View on HN

How leading edge will Intel's Germany fab be when it's finished? 10nm++?

While TSMC will be on what, 1nm by then?

Intel's PR link doesn't go into details on this.

Intel's "10nm+" (non-EUV) processors compete rather fiercely with AMD's "5nm" (EUV) processors. The latter are weirdly not 2x better, not even close, on any metric. Hm.

competes???? in what universe?

have looked at perf per W?

Anecdotal evidence, but I do know that my 12th gen Intel can burn through 10% of battery in 10 minutes. That is to say that Intel is doing really poorly in power consumption.

> power consumption

It really depends on the model. I see, for example, Kaby Lake Y models with a TDP of 4.5W . Though, in general, yes, many report that Intel is not taking power consumption as a priority (which does not mean that they do not offer niche "ultra/extremely-low power" products).

Will the renegade snipers make their point explicit.

I can't explain the downvotes, but for what it's worth both Apple and AMD offer far more power for the same TDP when talking about low-wattage parts, do they not? Thats the whole benefit of TSMC's process advantage (and also why Nvidia also moved to them from Samsung this generation). In practice of course, to get the performance uplift consumers expect, AMD and Intel and Nvidia are all pushing TDPs to the limit where they can, but for products that want performance-per-watt, the TSMC-based ones are king, from what I've seen.

That said, it's all a moving target, and anyone who thinks Intel isn't producing good parts today is wrong -- they are.

> AMD

I don't know: if you check graphs such as

https://www.cpubenchmark.net/power_performance.html

there is intersuccession of Intel and AMD. AMD seems to offer more models, but a proper analysis of their stand-outs and similarities is required to get a proper picture.

TDP is not a good metric to normalize against because it's a thermal design target only tangentially related to power limits. Power limits are generally much higher. For example the AMD stock limits result in a power target that is generally ~40 % higher than the specified TDP. Lots of motherboards will disable that limit automatically though and let the CPU draw as much power as it likes. I don't know how current AMD mobile platforms work, Intel Alder Lake has four or five different power limits. There's an instantaneous power limit, that's like 100 W. Then there's PL2, the short-term power limit (for a few seconds), which is like 30-40 W. PL1 is the sustained limit and depends on the laptop and is set to match the capabilities of the cooling system (or lack thereof) in an Intel-specified range (like 6-25 W). That's typically marketed as the TDP for laptops. There's an additional power limit for the iGPU and something somewhere is doing the divvying up of the currently available package power into iGPU power and CPU power (which doesn't seem to work that well under Linux).

In the current desktop generation AMD and Intel are much closer than in Intel 12th vs AMD Zen 3. That's mostly because 13th gen has pretty much the same limits as before, but is more efficient, while Zen 4 has massively increased the power targets throughout the stack - by 60 to 90 % - to keep up with Intel (plus the marketing advantage of getting through the 5 GHz wall)

E.g. a 5600X had a PPT of 76 W, a 7600X has a PPT of 142 W. A 5950X was 142 W, a 7950X is 230 W.

While they may be powerful, and they definitely are, they’re pushing out quite a bit more heat and requiring a decent amount more power to get there.

Though it’s impressive how much efficiency they’ve squeezed out of that node regardless.

But to say that TSMCs process doesn’t confer large advantages is silly. Intel even uses them for their new discrete graphics cards — they wouldn’t have been able to compete otherwise (among other less interesting business reasons)

It will be whatever the latest node is at completion, so most likely Intel 3 or Intel 20A [1]. It was Pat Gelsinger's pledge to get Intel's node roadmap back on track when he joined as CEO just under 2 years ago. From what they've been submitting papers wise to industry forums, IEEE Synopsium and the like, they're on track to deliver 20A by 2025. I recommend following the progress of Intel's new "Intel Foundry Services" business segment.

[1] https://www.intel.com/content/www/us/en/newsroom/news/intel-...

[2] https://www.granitefirm.com/blog/us/2021/12/28/tsmc-process-...

If Intel is not good enough for US, why would it be good for EU?

It's clear that EU did what US wanted instead of what would be in its interest.

Intel is making the right capex moves to position itself for a future situation where Taiwanese production could get interrupted. Intel is right alongside TSMC and Samsung at the leading edge and it would be an exaggeration to suggest otherwise. Marginal differences in road map and quality over the short term don't tell us the full picture in a business where course-corrections take years and years to bear fruit.

20B in Ohio: https://www.intel.com/content/www/us/en/newsroom/news/intel-...

20B in Arizona: https://www.intel.com/content/www/us/en/newsroom/news/intel-...

> If Intel is not good enough for US, why would it be good for EU?

Who said they're not good enough? They're slightly worse than TSMC, but still put out competitive chips for desktops, laptops and servers. Slightly more expensive, slightly higher power consumption but still, perfectly acceptable and still the market leader in some areas.

> It's clear that EU did what US wanted instead of what would be in its interest.

No, it's not clear. Why would the EU do that? They did the best they could given the very limited choice present.

Intel has lots of leading edge fabs in the US and continues to build them. They made some business mistakes around not opening up their chip design the way ARM enabled ... but the tech capacity is all there. Now they are doing contract production for chips that they did not design so I expect them to catch up in a few years.

Meanwhile Chris Miller in Chip War argues that Intel will fail again since nobody wants to give their secrets to Intel. If you’re Apple, for example, then handing over chip designs to Intel isn’t ideal in terms of competition.

Intel is still holds the market on laptops and desktops at about 3/4. They simply push more volume. TSMC makes desktop chips for Apple, but Apple isn’t the majority. The rest of TSMC’s production is almost exclusively GPUs and phones, and this could change. If Intel executes well on their next nodes, and should TSMC slip at all, Intel could pull ahead. Never count Intel out. People thought Intel would die after the z80 ate their lunch, after the Athlon 64 ate their lunch, and now they say it again… I think it’s all just noise.

The market isn't laptops/desktops and phones, everything has ICs in it these days. Laptops/desktops and phones are just the high-margin items.

The blast-radius for non-intel chip fabricators is much wider and impactful for the economy

Sure, but most of the rest isn’t leading edge and more manufacturers are in that space, including TSMC and Intel.

I always find it odd that companies always want to be build things outside of their home state

TSMC -> US

Intel -> EU

Toyota -> US

Honda -> US

Ford -> MX / Canada

GM -> MX / Canada

The most common reason is typically cheaper labour. That’s why Nike produced in Japan and then moved to even cheaper countries. Related might be that some countries have a lot of knowledge workers in their 40-50s who can then manage factories in other countries. Japanse companies producing cars in Europe is an example of that.

For the current trend to move to the US. That has to be government funding. Luckily, fabs require large investments at the start which might break-even more expensive labor. Also, maybe some more automation in the fabs could also help in making the fabs cost competitive.

>>typically cheaper labour.

While a factor, labor costs alone are the primary reason. In reality is more about the wider regulatory burden on a company, including environmental controls

That said it does not explain why a Japan Company can make cars profitably in the US but a US Manufacturer can not.

If it was only labor why would Toyota not have a factory right next to GM's in Mexico and import cars from Mexico to the US?

I guess its that the Japanese Company can make cars profitably anywhere but making cars for the US market in the US means there's no/less import taxes, also because the factories are huge they often get local tax breaks, and the cars may also eligible for EV/hybrid tax credits (I think Toyota have sold over 200K so there cars can no longer get this).

I think making cars in Mexico for the US became less worthwhile when Trump pulled the US out of NAFTA.

>US became less worthwhile when Trump pulled the US out of NAFTA.

lol, trump did not "pull the US Out of NAFTA" not only is that not with in the power of the president to do, it was never on the table, NATA was replaced with USMCA, and only made a few small changes, for Automotive that means 75% of the vehicle components must be made in MX, US, or Canada, up from 64% under NAFTA

Nothing in the law would impact making at car in the US vs MX, and was aimed at preventing increased parts from China or other non-north American nations, and all of the changes were passed by congress with wide bipartisan support

OK - but he signed a deal with Mexico that replaced it and didn't the new deal mean that more parts had to be from the US and half the car factory workers needed to pay $16 an hour - which would make Mexico less attractive?

> didn't the new deal mean that more parts had to be from the US and half the car factory workers needed to pay $16 an hour

Both were proposed changes that were not adopted in the final version.

I thought Japanese car factories in the US were all non-union (as opposed to the US brands)? Toyota does have a Mexico factory at least according to Wikipedia.

Correct, the Honda/Toyota plants in the USA are non-union.

Interestingly enough the “Clean Energy for America” bill pushes for additional tax incentives on EV's built by companies using union labor:

More specifically, the proposal says that electric vehicles assembled in the United States would qualify for a $10,000 tax credit while EVs that are built at facilities whose production workers are members of, or represented by, a labor union would be eligible for the full $12,500 credit.

Aren't clauses like this just a clear cut sign of corruption?

It doesn't even make a lot of sense as an incentive to have workers unionize, since why would workers care what price what they're selling cars for. The law would only exist to benefit existing unions.

> Toyota -> US

you dont pay tariffs if you produce locally. not hard to understand.

Toyota builds trucks in the US because of the chicken tax.

https://en.m.wikipedia.org/wiki/Chicken_tax

US-Japan trade war was a big thing, and Japan relies on US for national security. There was no option not to manufacturing in the US.

> and Japan relies on US for national security

Good luck with that bet in the future.

It's not that odd that international companies would build local facilities for major markets they're selling into. Toyota sold cars to the US, and building plants here made sense relative to that for a variety of economic and political reasons. Everyone makes more money together so everyone's happy with the arrangement.

I find it more odd to presume that corporations have some nationalistic competitive interest.

Cars are big and heavy, and so building them near the buyer has a logistics benefit. Chips have higher value per weight and size by orders of magnitude.

Chips are strategically valuable, would the EU back a competitor to have local manufacturing capacity?

Three reasons:

1. By operating a local subsidiary, you avoid unfavorable regulation and sometimes protectionism (Toyota/Honda/Intel).

2. Cost savings (transportation, labor, taxes, etc).

3.Proximity to customers.

These are all international companies. Ford of Canada builds in Canada. GM of Canada builds in Canada. Honda US builds in the US.

It's incredibly provincial to think of multinational companies so simply.

Reply on news.ycombinator.com