Three-Way Deadlock: Part II
The Battle Between NVIDIA, Intel and AMD
As we continue to part two of this series, Three-Way Deadlock, we will take a look at the most established company in the semiconductor industry, Intel. Intel is no stranger to us. For a long time, almost every desktop PC, laptop and data center had an Intel CPU running it, and is still true today. According to Tom’s Hardware, Intel’s share in the desktop market, laptop/mobile market and server market is 80.8%, 80.1% and 94.2%, respectively, as of the second quarter of the year.
However, as we see in the chart above, it seems that the trend for these numbers has been consistently dropping. Is this the end of the story for Intel? Or will it be able to see itself bleeding and stop it before it’s too late?
Intel is not just simply a CPU or a data center product, so to give it a better context, it would be important to take a look at the Intel x86 architecture and differentiate it from the ARM architecture.
Intel designs and manufactures its own chips, while ARM only designs the CPU and GPU cores then it licenses them to various companies.
The architecture for both is different. Intel uses the complex instruction set computer (CISC) architecture, while ARM uses the reduced instruction set computer (RISC) architecture. Since the CISC requires extra logic to perform complex, multi-stage instructions, it would demand more power, thus increasing the electrical usage.
Photo taken from https://www.ahirlabs.com/difference/cisc-vs-risc/
The manufacturing stage for both is also different. Intel’s foundry is currently on 14nm/10nm wafers while most of the ARM clients who go to Samsung or TSM’s foundries are on 5nm/7nm wafers, albeit that doesn’t mean that ARM’s clients can’t go to Intel’s foundries for manufacturing.
With that out of the way, let’s move on to why defining this matters.
Photo taken from Intel
Clearing the Air
There is a lot of misconception out there when it comes to the chip sizes that Intel is putting out compared to the chips from TSM or Samsung. Just because the 7nm is a ‘smaller node’ doesn’t mean that it is actually more advanced than the 10nm.
Data taken from https://en.wikipedia.org/wiki/10_nm_process and https://en.wikipedia.org/wiki/7_nm_process
In reality, the differences in their processor technology aren’t so far apart from each other.
Rough End to the Old Decade, Rough Start to the New Decade
The latter half of the past decade was a struggle for Intel to move past their 14nm processor to 10nm. Intel has delayed the production of its 10nm processor until late 2019 to this year. Just when we thought that they’re back on track with the 7nm on the way next year, their CEO, Bob Swan, announced during its earnings call for the second quarter of 2020 that they will be delaying the production and launch of the 7nm to late 2022 or early 2023.
Cost of the Delays
If Intel’s processors aren’t necessarily lagging behind TSM or Samsung’s ‘lower-numbered x nm’ processors, then the delays shouldn’t really set Intel back, right?
Here’s where defining the two different processor architectures came in handy. Since Intel’s x86 would typically require more power, having smaller transistors would provide better electrical efficiency and the ability to have more computing power. From large data centers to laptops, all of them want better performance and better power efficiency, because, for their manufacturers, this would mean lower costs, higher margins, or even both. Therefore by being able to roll out the 7nm sooner and go further down the ‘lower-numbered x nm’ processors, Intel’s processors would be certain to regain market share.
But that doesn't seem to be the case in the near future.
As we start seeing NVIDIA getting ready to overtake the data center space with its new DPUs and AMD teasing the market with its new generation of Ryzen CPUs, Zen 3, while Intel is constantly delaying itself, Intel will surely experience a much deeper cut into the market share that they hold.
The problem here is that, for one, the delays in the production of the 10nm processor didn’t really budge Intel’s revenues; in fact, it even grew by 14.5% from 2017 to 2019.
This is why they might’ve initially thought that this slippage would bear minimal consequences if any. But alas, during the earnings call for the third quarter, Swan mentions that fourth quarter revenues are projected to be down 14% year-over-year, with the PC segment down single digits while the data center segment down 25%. And I think this all boils down to Intel’s complacency for being on the throne for too long.
So when a large company like Apple announced last June that they are replacing the Intel processors on their Mac line with its own ARM-powered processors, you know that this might be the spark that leads to Intel’s demise. Customers of Intel will slowly realize the incentive of moving away from its x86 architecture, and I am convinced of this as a former principal engineer and performance architect of Intel, François Piednoël, shed light as to why Apple moved to ARM, he said:
“The quality assurance of Skylake was more than a problem, it was abnormally bad. We were getting way too much citing for little things inside Skylake, and Apple became the number one filer of problems in the architecture. And that went really, really bad. When your customer starts finding almost as much bugs as you found yourself, you’re not leading into the right place.”
With this revelation, Intel has really got to get a handle on their situation and realign themselves with what they really want to do in this industry. Because right now, there is every reason to doubt Intel’s ability to deliver.
Takeaway
I wouldn’t exactly say that Intel’s story is about to end, at least not in the short-term, with their deep pockets, they might just be able to put themselves back in the race again. However, with Swan using the gains from the recent sale of their NAND (flash memory) business prioritizing share buybacks over reinvesting capital to R&D, plus the fact that competition is hitting them left and right in its PC business and data center business, it will be hard to consider Intel as anything close to an investment. We’ll see along the way whether or not Intel will be able to pull through, and if they do, then maybe I’ll think about rewriting my story for Intel.
Note: I just want to inform the reader that there may be some technical aspects that were overlooked primarily because my knowledge on machines and semiconductors itself is very limited. So should you find any mistakes or parts that are lacking in material information, please do inform me.