That's only a partial explanation. There has been FAR less competition in the CPU space than there is in the GPU space in the last 5 years. MicroCenter has i7-6700K for $259.99 and i7-6800K for $329.99,
both prices which is lower after adjusted for inflation than the MSRP of
January 2011 i7-2600K. If Intel priced i7-6700K at $634 (double the price of i7-2600K), I would simply refuse to buy it. Lack of competition alone doesn't explain the willingness of so many consumers to pay double for the same historical tiers. There is a lot more to it like prices of high-end audiophile headphones have increased 2-3X in the last 5 years despite the
most fierce competition.
When HiFiMan sold HE-300 for
$249, we now know they still make a profit at
$99. We also know that Sennheiser's HD650s which debuted at
$550 are profitable even at
$199.
With NV, what they are doing is following the foot-path of the overpriced audiophile industry, and sprinkling it with Apple marketing, creating an emotional attachment and desirability of the product.
Had Titan XP sold for only $499 like the GTX580, it would have never been as desirable to PC gamers and as a result the $379-449 GTX1070 and $699 GTX1080 wouldn't have looked like such an amazing value relative to the Titan XP. NV still hasn't matched Apple though as Apple charges $200 for an upgrade form an i5 to an outdated architecture generation i7 in the iMac when the entire i7-6700K costs $260 at MC. Oh, let's not forget $200 for 8GB of DDR4 memory upgrade. NV still has room to raise prices even more during Volta as Pascal is wildly successful despite
record GPU prices per each GPU tier. Just goes to show that PC gamers are willing to pay even more. If mid-range die 1080 FE sold like hot cakes for $700, a $900 1080Ti will be instantly sold out by
the same people who are about to dump GTX1080s on the used market. That's the missing link in NV's strategy -- since next series makes the old series obsolete and thus lose a huge amount of resale value, it will make it too costly to buy $700 cards and NOT upgrade to the next series. This gives NV leverage to price 1080Ti/2080Ti above GTX1080.
Do you know why Intels margins have no shrunk(actually growing over the years), it because they are using smaller and smaller dies because they are keeping their core counts the same and they are not increasing pricing because the performance jump is not there and they are selling you alot smaller die each time.
Nvidia does makes it more transparent with its more model numbers, Intel is doing the same if not worse.
The 6700k die size is 122mm2, the 4700k was 177m2, sandybridge was 216mm2 and lynfield was 296mm2.
And with intel v2011 platform, Intel has been selling you a chip with a more and more disabled die.
The 980x which was the first 6 core intel consumer chip equivalent xeon chip the 5600s also had a maximum of 6 cores. This means you were getting the whole die.
Starting with sandybridge, the 6 core chip actually came from a 8 core xeon chip e5 4600.
Ivybridge the 6 cores came from a 12 core chip xeon. E7-8xxx v2
Haswell e a 6 core from a 18 core xeon 2600v3
The same is with broadwell. A cut down from an 18 core processor.
What intel has been doing to decrease cost on their 6 cores is giving you a more and more disabled chip. We might be paying 400 dollars for a 6 core die compared to the 600 and 1000 we used to pay, but we are only getting a third of the cores enabled.
Atleast with Nvidia high end chips, you getting something that resembles the full die size for the price.
A titan much more resembles the full die than Intels extreme editions.
What Intel has been mostly doing is giving you the same amount of cores which means they are using a smaller die or a chip with more cores disabled. This has resulted in the terrible performance jump between generations and a smaller product that is cheaper to produce and hence why prices don't need to increase.
With Nvidia, your getting more cores, but your paying for it, but your also getting a better performance jump.
If Nvidia played the intel with no cost between generations, pascal might have the same core count as the the gtx 980 or the gtx 980 might of had the same core count as the gtx 680, just leveraging architectural improvements to sell cards.
Both are greedy, but as a tech enthusiast, i prefer the later because we atleast get progress, and Nvidia jumps in price in performance have been much higher than Intels between generations.
Thus compared to Intel, Nvidia is the less greedy of two monopoly models.