Of course, our well-placed sources want to remain unnamed but Intel’s 10nm delay was for a financial reason and not the inability to make a good 10nm CPU. Some might remember that Intel has many fabs and that one of the main reason why Intel got so successful in the CPU business. Managers of the fabs are good with the fab economics.
Our own Mike Magee posted that Jerry Sanders III, the former CEO of AMD, said at a conference a few years before 1999 "only real men have fabs".
Intel's CPU group is ordering wafers from Intel fabs and despite the fact that from the outside it looks like the same entity, they both act as two separate companies. Intel fab wants to make money on the wafer whenever it can.
A 300-mm wafer can cost between $200 and $700 and let’s assume that the new 10nm wafer could easily cost close to $700. Since Intel aims for very high yields, sometimes up to 90 percent, they want most of the 10nm dies to work. Let’s assume that the next gen 10nm Ice Lake core is 15x10mm (150mm2). This would mean that Intel can get approximately 394 dies per wafer. Now the yield is the number of how many of these dies will work. The higher the number, the better the economics.
AMD and GlobalFoundries had products that had less than 50 percent yields, or even less than that, but still went into production. our sources close to the matter explained that Intel doesn’t want to move to a new node until it gets very high yields.
So, when Intel starts shipping its 10nm, with Cannon lake and Ice lake in the course of 2018, Intel aims to have very high yields.
The fact that AMD didn’t have anything to push Intel for more than a decade made it lazy, but Intel doesn’t like the fact that Samsung and Qualcomm have managed to move to 10nm faster than it did. Again, bear in mind that these are apples and oranges as Intel makes much bigger chips with 5 to 120W TDP, while Qualcomm and Samsung aim for 2.5W TDP chips.
Qualcomm went very public saying that its Snapdragon 835 in 10nm was only 72.3 mm - a huge shrink from Snapdragon 820 with 113.7 mm2, but again these are two totally different markets. Intel U 14nm notebook series have 15W TDP while the H series has 45W and the desktop goes from 65 to 95W.
Since Intel didn’t havw any competition from AMD in the desktop and notebook market until the middle of this year with Threadripper and Ryzen/Epyc processors, Intel was fine rehashing its 14nm product and essentially going for 6 core desktop and 4 core notebook products with Coffee Lake S and Kaby Lake H, when necessary. Intel already refreshed its Kaby Lake R 8th generation notebook core processors with 15 to 25W configurable TDP that managed to get better battery life and significant performance over the dual core 7th generation. The core count jumped from two to four without hurting battery life which is simply impressive.
Now, imagine these cores in 10nm, in the course of 2018.