Intel isn’t in a very comfortable position at the moment. From having a monopoly over the CPU market just a few years back, the company is now struggling to meet 14nm chip demands, all the while AMD is rapidly regaining lost territory. The server market which used to be an Intel-only space is also being encroached by AMD’s 64 core Epyc Rome behemoths. In the coming weeks, with the launch of the Ryzen 4000 “Renoir” based laptops, the same is expected to happen in the mobility and OEM space.
How did it come to this? Well, as far as the 14nm shortage is concerned, there are many reasons. Unforeseen chip demand, unexpected product launches, ramping up of 10nm parts and so on. However, there is one reason for Intel’s supply shortages that no one talks about. The AMD Ryzen effect. Let’s have a look.
It all Started with Zen
In 2017, when AMD’s 1st Gen Ryzen first debuted, Intel was facing a highly competitive lineup from AMD the first time in over a decade. This was in the midst of when the company was trying to migrate to the 10nm node (back then Canon Lake). Intel was literally caught with their pants down.
After the 1st wave of the 10nm Cannon Lake parts turned out to be a disaster, Intel double-backed to the 14nm node, producing three back to back generations based on the Skylake core (and a fifth one in the form of Comet Lake). These were Kaby Lake, Coffee Lake, and the recent Coffee Lake refresh. They shared the same microarchitecture and process node as Skylake, and the only difference was that the core counts increased with each successive generation. This was done to keep up with AMD’s Ryzen CPUs: The Ryzen 7 1700X offered 8 cores and 16 threads while the competing Intel part, the Core i7-7700K was a quad chip.
The 8th Gen Coffee Lake lineup increased the core count of the Core i7-8700K to six while the 9th Gen Core i7-9700K had 8 cores and 16 threads. The upcoming 10th Gen Comet Lake-S lineup will include as many as 10 cores for the Core i9-10900K. As the process node and architecture were the same, this meant that the die size of Intel’s CPUs increased by almost 40% from the 7th to the 9th Generation.
The die sizes of Intel CPUs over the last few generations (Sq. mm):
- Kaby Lake: 126
- Coffee Lake: 152
- Coffee Lake Refresh: 178
As the die size grew, the number of chips produced from a single yield decreased. Basically, as the wafer size or density didn’t increase, the number of CPUs obtained from each wafer decreased significantly. However, market demand increased. This meant that within just a couple of years, Intel had to increase its 14nm production by more than 50% to avoid shortages. Only, that didn’t happen.
The company took multiple steps to expand 14nm wafer production, but it just wasn’t enough, and the results are quite apparent today. After more than a year of shortages, Intel is forced to deal with a highly resurgent AMD and disgruntled OEMs shifting to AMD’s Ryzen CPUs to meet demand, simultaneously.
10nm Ice Lake, Xe, and 14nm Foundries
Of course, Intel has been investing billions in new foundries worldwide to improve 14nm production, but that’s a gradual and painstakingly slow process. By the time Kaby Lake production was over, the 14nm yields must have been quite high but when you suddenly increase production by one-half, there are bound to be problems. The new production lines are starting from scratch and don’t produce the same yields as the original foundries.
This means that nearly half of the yields are sub-par or at least worse than the existing 14nm capacity. This only makes matters worse. One of the reasons we’re seeing so many F CPUs (Core i3-9100F, 9400F) is because Intel can’t afford to discard chips with faulty iGPUs as the production is already falling short as it is.
The reason why these F parts aren’t that much cheaper than the non-F variants is cos that would increase the demand for them. And Intel needs to keep it as low as possible. The 14nm yields will be back to normal soon enough and when that happens, the company will have to disable chips with working iGPUs to serve the F market, essentially selling them for a loss. That is the reason why the F parts cost nearly the same as the non-F parts.
Then there’s the new 10nm Ice Lake and Xe GPUs waiting volume production. Although Intel finally managed to turn the 10nm node (10nm+ in this case) into usable Ice Lake chips, they are still absent from the volume markets.
Intel has promised 28 core Ice Lake SPs by mid-2020 as well as the Xe GPUs by the end of the year. Since this is going to be another monolithic design, it’ll put a significant strain on Intel’s foundries especially the 28-core server parts.
Although Intel’s 10nm node failure can be attributed to an overly aggressive policy or perhaps the death of Moore’s Law, the 14nm shortages are a different case altogether. Intel didn’t anticipate that AMD’s Ryzen CPUs would be as disruptive as they turned out to be. And they certainly didn’t expect that their 14nm offerings would fail to match up to the 3rd Gen Ryzen parts even after doubling the die size. In the end, this reminds me of the tale of “The Hare and the Tortoise“.