Friday, August 5, 2022

AMD vs. Intel

Dan Luu:

Looks like AMD passed Intel in market cap last Friday, after being fairly close for quite a while.

The majority of comments I’ve seen are betting on AMD, but I’d bet, at even odds, ten years from today, the 1-month trailing average market cap of Intel is higher than AMD’s.


I think Intel will be ok if it can recover to 2010-levels of dysfunction while it’s much larger than AMD in revenue/scale.

Ben Thompson:

While there are a host of reasons why TSMC took the performance crown from Intel over the last five years, a major factor is scale: TSMC was making so many chips that it had the money and motivation to invest in Moore’s Law.

The most important decision was shifting to extreme ultraviolet lithography at a time when Intel thought it was much too expensive and difficult to implement; TSMC, backed by Apple’s commitment to buy the best chips it could make, committed to EUV in 2014, and delivered the first EUV-derived chips in 2019 for the iPhone.


Time will tell if the CHIPS Act achieves its intended goals; the final version did, as I hoped, explicitly limit investment by recipients in China, which is already leading chip makers to rethink their investments. That this is warping the chip market is, in fact, the point: the structure of technology drives inexorably towards the most economically efficient outcomes, but the ultimate end state will increasingly be a matter of politics.

See also: Dithering.

9 Comments RSS · Twitter

This is not new. Intel knows how to produce 3 nm and even smaller chips, and has known it for decades now. But they are delaying it more and more, and more and more.

Why? Because once the limit is reached (quantum effects at about 1 nm), the business is almost over. So, instead of reaching such end in 10 years, their plan is to do it in 100 years.

How? Promising and delaying, again, gain and again. They are experts on that.

They are also experts on promising and not delivering. Remember the Intel Optane (Xpoint) NAND fiasco.

Nothing new. The amazing thing is that people have not noticed that decades ago.

Old Unix Geek

@MeX: no, Intel does not know how to make 3nm chips, nor do they have the lithographic machines to do so.

Old Unix Geek

The stratechery article contains an error: SMIC has been shipping chips with 7nm features using a DUV process for over a year.

"But they are delaying it more and more, and more and more."

If that conspiracy theory ever had any merit (which it did not), it lost it when Zen 3 came out, completely destroyed Intel's offerings, and they were incapable of responding in any meaningful way.

AMD's market cap has now surpassed Intel's. If Intel has a working 3 nm process, Lisa Su must be paying Pat Gelsinger billions not to use it.

To "Old Unix Geek" and "Plume":

Intel is following the described strategy because it works for then. Because people follow inertia.

As examples, people used the horrible command-line DOS when the wonderful Mac GUI was available. People used and continue using the awkward Windows when Mac is available. And people use Intel x86 when ARM is available.

Inertia. It is inertia. And intel knows it. Remember that when they reach 1 nm or so (quantum effects) the big business will be over.

And finally, why did intel reduce the lithographic node just a little bit to increase performance about 20 to 30% every 18 months or so, again, again and again over decades? Because they did not know how to reduce it even more? Never ever? Obviously, not. They just want to milk the cow as much as possible. And it works for them because people seem blind about it. Really amazing.

Old Unix Geek

MeX, it's pretty obvious you are speaking about things you do not understand.

In the real world, progress stops if you stop investing in it, which is what the business people at Intel did. The optical limits of the lithographic process are getting harder and harder to master increasing the costs of such development. If you wish to debate, and prove your competence in this area, please explain precisely what limits we are dealing with. And no, it has nothing to do with 'inertia' or 'quantum effects'.

"Intel is following the described strategy because it works for then"

It hasn't worked for them. They just posted a 500 million loss, they've had to delay their server CPUs while AMD's server revenue has almost doubled this year, and they're also rapidly losing grounds in the consumer market. They went from holding 92% of the x86 notebook market share to holding 73% in Q1 2022, with the trend line pointing down at a worryingly steep angle.

FFS, they have to fab their new GPUs at TSMC just so they're even vaguely in the same performance/power ballpark as Nvidia and AMD.

They're running out of money so quickly that they're now forced to kill products like Optane.

Their R&D budget flattened out around 2013 while they added lots of new products. They've coasted on their earlier success for the past decade, and now they're paying the price.

They're not clever. They don't have an ace up their sleeve that they can pull out at any time. They're screwed.

To "Old Unix Geek" and "Plume":

I agree with some things that you say, but not with others. As an example and said, Intel has been delaying releases of much better microprocessors for decades, just increasing performance about 20 to 30% only every 18 months or so. And many times promising and not delivering, delaying even more. I can understand that they might have problems at a particular moment, but not always. You seem to consider that Intel could never release much better microprocessors in all its history, for decades. No, that is simply not true. It was clearly planned.

It is not just a problem of lack of investment. It is also a problem of programmed delay and lies. How many times has Intel promised things (just to keep and increase shareholders) and then delayed them again, and again, and again, or even not delivering at all, clearly cheating, like in the case of Optane?

Intel might be losing money now due to competition, but they choose to continue with their delays and false promises because as said, once quantum limits are reached, the game is almost over.

In summary, it is a mixture of factors. Some of your points are true, but also some of mine. The final conclusion is that Intel could have done much better in the past and present for sure, but they decided otherwise. And worse of all, it seems that they have also decided to continue with such strategies. Fortunately, there is competition, like TSMC.

"How many times has Intel promised things (just to keep and increase shareholders) and then delayed them again, and again, and again, or even not delivering at all, clearly cheating, like in the case of Optane?"

You seem to imply that this is a point in your favor, but I think it's not. Intel clearly kept overestimating their own capabilities, and then failed to deliver again and again. That's not evidence of a ruthless company intentionally delaying things to keep the gravy train rolling. If it were, they wouldn't needlessly announce things they have no intention of delivering, only to end up making themselves look incompetent.

Preannouncing things you can't deliver only counts as FUD when it works to your benefit (i.e. when it kills a competitor's market because people wait for your product); in Intel's case, it counts as the Osborne effect, because it drastically shifted market share to Intel's competition.

This is evidence of a dysfunctional corporation, where people along the management chain keep lying to each other to validate each other's predetermined conclusions.

As a general rule, stories that assign some kind of successful long-term Machiavellian strategy to corporations are almost always wrong, because most people are dumb, and most CEOs are people.

Leave a Comment