The Intel Core DUO 2 CPU had a ‘solo’ variant which was off the same assembly line. Pricing is indeed a dilemma.
Very early in the Soviet development of micro processors the few to get such a processor frequently found it accompanied with a short list of instructions that that particular chip was unable to execute.
Gene Amdahl’s company Trillium was predicated on a large chip to be a 370. It was to have redundant CPU elements to increase yield. I don’t know if they planned to deliver chips of various performance from the same line. The notion of redundant cache sections was already proven.
The day that I visited the SWAC computer at UCLA in perhaps 1950, the machine was not running. It spent most of its time in this state as they were busy ‘improving’ the architecture of the machine. The number of bits in a word varied from day to day, depending on how many Williams tubes were working. There was not much running code at that point.
In that short interval (1959) between the invention of linked lists and the invention of Hamming codes, I and several others contemplated programming machines with enormous memories with known bad spots. The need for contiguous good words to hold arrays might be thus avoided. Today RAM and disks ‘transparently’ hide bad spots.
Ted Hoff tells of a DAC part that Intel made where critical resistances on the chip depended which position on the wafer the chip was born.
Warning: The following story is 3rd or 4th hand:
Seymour Cray visited an early transistor manufacturer. They showed him how they tested the transistors, for linearity, I think, rejecting many for failing the test. I suppose that early customers were building analog circuits. The next day Seymour asked them if he could buy a batch of rejects, and then went on to build perhaps his first transistor computer, perhaps the CDC 160. Many of the reject transistors suited digital circuits better than their linear brethren.