Another way of looking at it - which I believe was not covered in the article - is observing chip areas. If you look, top-of-the-line server chips (Xeons, etc.

Another way of looking at it - which I believe was not covered in the article - is observing chip areas. If you look, top-of-the-line server chips (Xeons, etc.) pretty consistently hit 300 mm2. On the other hand, mid-range desktop chips usually are about 100 mm2. Besides marketing, pricing also has to do with process yield, which is not linear: 3x area is more than 3x more expensive. What to do with that area? Given the shrinkage that allows multiple CPU cores, the answer that gives best price-performance is cache, after the first 4 or 6 CPU cores ...