It is a truly complex story that requires an in-depth study of the differences offered between GPU and CPU
The size of the wafer and chip determine
As we mentioned, size is the key to wafers and it is a chip thing. The high-end chip usually features a much higher number of transistors than the rest of the downstream range. Therefore, larger number of transistors means larger space used by each chip, ergo means the minimum amount of chip per unit.
Assuming the success rate is excessive 70% of valid chips, the cost of the fish is certainly a profit for the chip manufacturer when it comes to selling its product through a standard or server market.
But manufacturers have to calculate these sales based on final product price and demand, so ordering lifers as the market is a risky move, especially given the success rates ahead of EUV. So, the big three have been making a strategy to follow: increasing the number of chips regardless of the success of the wafer is different matrices
This means that in recording the wafer will be used for a particular matrix that will incorporate various chips into the market, where the cutting of some and others will depend on the degree of its success, always referring to the chip with the perfect markings.
That is, if a particular chip does not comply with the high-level provisioning, it will move to a lower level with lower performance, be it speed, cores, cache or architectural components such as GPU or memory controllers. This is true to some extent, because if the features are very different it is of great benefit to create another small matrix and increase the number of chips successfully in each cloud.
A smaller chip has an added benefit, as many more are created and the success rate increases by percentage.
Is a high end more expensive than a low end versus their chips?
The fact is that yes, and as improvements in architecture are created the gap grows more precisely because the lithograph processes have been temporarily suspended.
The arrival of EUV for large centers and companies it is a temporary relief that should simplify and reduce the cost of production per wafer and chip, because smaller patterns are required per recording, the success rate per chip will be higher and with this product per product it will cost more to manufacturers.
This does not need to be reflected in the final prices of processors and graphics cards, because an investment in R&D or a jump in yield may cause the final consumer price to remain.
What is clear is that a high-quality chip, whether CPU or GPU, to get the physical space of the chest, will be more expensive than the low end, sometimes with an undivided number between them. The other issue will be that, in fact, this cost is justifiable in one and the other grade as the final product, with the entry of other classifiers that need attention.