Intel is a Victim of its Own Arrogance
innovationnation.blog
innovationnation.blog
There is also a huge political headwind when it comes to onshoring leading edge semiconductor fabrication - no company will benefit from this more than Intel.
I’d expect more pain in the medium term - they certainly missed the AI boat - but I wouldn’t be surprised if Gelsinger succeeds in turning things around, and I’m rooting for them. Competition between AMD and Intel, and between TSMC and Intel - will be nothing but favorable for core count, $/core, and the continuation of Moore’s Law.
Here is an example of the 144 "E"core Sierra Forest matching performance of the 60 core Sapphire Rapids 8490H in the phoronix benchmark suite: https://www.phoronix.com/review/intel-xeon-6700e-ampere-altr...
Sierra Forest does better if perf/watt (or energy consumed per run) is the metric, but still loses to the 64 core Epyc Siena in those same benchmarks.
I would guess there are some very efficiently parallel workloads where tons of "E" cores win in performance, but the current Intel CPUs do not seem to be competitive in my non-expert opinion.
"As evidenced by Apple’s M series chips, ARM has taken a huge bite out of x86’s market share. While ARM was focused on power efficiency, Intel was trying to brute force power out of older designs by throwing more watts into every successive chip. In the massive laptop and data center segments, this is a losing strategy. Not only that, but they significantly missed out on EUV by not investing a decade ago. Now the combination of TSMC, AMD, and ARM are slowly but surely eating away at Intel’s market share. All 3 companies are worth significantly more than Intel at this point."
Citation needed? I don't see many desktops with ARM chips. Most of the servers I see are still x86. etc.
I'm no Intel apologist, but Intel is dealing with the standard problem that investors love fabless semiconductor companies and hate actual fabs.
And everybody seems to have forgotten what happens when fab capacity is constrained and your fabless companies can't ship a single thing. A strong earthquake or tsunami that affects Taiwan and all those fabless companies shut down again. But, then, goldfish have longer memories than business executives.
Neverthless, Intel does have structural problems. Technically, the problem is that Intel is a fab first and foremost and everybody else are second-class. This means that really good design engineers and software people are going to be hamstrung in the company.
Politically, Intel is a company in which everybody stabs one another. This is normal for most big companies, but Grove actuively encouraged it. Consequently, people hoarded their knowledge as it was valuable in a political firefight and now that lots of them are retiring/retired that knowledge got lost permanently.
There are millions of new devices that require processors every year, and if you're NOT making ARM processors then your fab is not making as much money as the competition. Your competition is going to have bigger budgets, and their engineers and fabricators are going to have much better high production experience.
Intel has let themselves fall behind. They've sacrificed the ability to stay competitive in order to stay loyal to their in house and very legacy designs. Intel is trying to sell the same product line that they've had for the past 15 years like the market didn't change, and everyone else has moved on. Now they have this 13th and 14th generation bug/defect, and they're going to look really unappealing.
> if you're NOT making ARM processors then your fab is not making as much money as the competition.
Nope. The profit margin on ARM processors is at least an order of magnitude less than x86, probably close to two orders of magnitude. If you are making x86 profit margins, making ARM processors is business suicide. This is true for both Intel and AMD.
Which applies equally to Apple devices, if you're watching the sales. They followed the same trend of moving units during COVID and then relaxed upgrades afterwards.
> if you're NOT making ARM processors then your fab is not making as much money as the competition
Do you know how ARM works? Or even chip production, for that matter?
For one, foundries don't choose to make ARM chips, their customers do. A foundry can and will make the same amount of money as their competition investing away from ARM, as Nvidia has shown on TSMC 4nm. Making "as much money as the competition" only works if you don't have to pay ARM license fees, and Apple is just about the only company in the world with that level of agreement with Softbank. Everyone else is avoiding ARM like the plague for it's dogshit licensing terms and 2-year OEM support contracts. The actual "deal" you get from buying ARM is the choice between their godawful off-the-shelf cores or the ISA itself. There are vanishingly few companies that want either of those things, least of all with ARM Ltd as the middleman.
For two, ARM's "golden era" is behind us now. If you've worked in electronics you're probably well familiar with the story of the ESP-32, a wildly successful microcontroller that hasn't been "officially licensed" ARM in years. The rare few ARM chips that are profitable don't respect the ISA license they're supposed to pay for, leaving only Apple and Nvidia as legitimate customers. Of them, Nvidia has been shipping RISC-V in place of ARM for over a decade and Apple would too if they had to pay for the IP. ARM is desperate to find customers that don't want to take advantage of them, but cannot find a single one.
Cloud providers used to buy mostly Intel servers, but now have various efforts on buying or making their own CPUs. Amazon is now at the 4th gen graviton, seems nearly inevitable that other cloud providers will follow. Microsoft is well on it's way with the azure cobalt 100. Google has the Axion.
https://news.ycombinator.com/item?id=33529838
Intel has a bigger problem because they are building fabs and chips but they’ve been working on their new plan for a few years