Okay guys here's the deal:
Nvidia actually wants to price their big dies (500mm^2+) higher than $500. This is because they're very expensive, have a lot of R&D money put into them and have low yields.
What made the 480 and 580 cost $500 then? Competition.
Let's look at the circumstances at the launch of the past pew NV flagships:
GTX 280: Huge die, priced at $650 with basic high end GPU features, has to drop in price very soon because of AMD releasing a GTX 260 matching 4870 for ~$300. NV has to stay competitive and was relying on their high end GPU for the mainstream stuff as well, they have to lower the price in order to sell it because G92 isn't cutting it anymore.
GTX 285: Same story but without the initial high price because of AMD already having their cards out. Only shrinking the process and a bit higher OCs and better thermals aren't enough to price it much above the then discounted 280. NV is still using G92 for their lower end parts and has to still manage some of the mainstream ($300 range) with the expensive GT200b. GT200b did though save them from making a nearly 600mm^2 die for mainstream cards which is probably the reason for why it was even die shrinked. The price can't go higher because of AMD's 4890.
GTX 480: Again a massive die, card is massively late, hogs more power than ever seen before (single GPU) and has lots of bad publicity even before launch. The performance barely warrants the $500 price compared to the $400 5870. NV tries to market the card with more VRAM, CUDA, PhysX etc. They also have to use the die for 470s and 465s again making it a mainstream chip as well. This isn't a choice, it's a must for them because of AMD's product range. The smaller GF104 appears much later and is no match for AMD's products and can't be used to counter high end products.
GTX 580: NV is unaware of what AMD might release after seeing the 5870 matching 6870 (well almost). The 580 provides 10% increase over the older 480 while actually decreasing the die just a tad, improving thermals etc. They can't price it much higher than their own card because the 400 series wasn't even selling well and the 5870 had been discounted to around $370. a $500 price is set as a counter to possible future AMD GPUs, however the 6970 that was originally planned as a $500 card (leaks showing listings from different etailers etc. showed this) barely matches the 480 and can't compete with the 580, AMD has to price the card similarly to the discounted 5870.
At this point NV saw how important the GF104, GF114, GK104 die was due to the huge volume of sales they produced. A decision is made to give the mainstream chip more attention so it will be on the market sooner.
GTX 680: This time NV already knows the performance of the 7970 which has launched at an expensive $550. Multiple leaks originating from before 7970 era launch plans of NV (drivers, early cooler images etc) show GK104 being introduced as GTX670Ti due to NV thinking it can't match a single 7970. After seeing how AMD's new card performs and where it is priced though NV can bump the name of GK104 up to GTX 680 and grab a big portion of the high end GPU market by releasing it at $500. GK100 is either failed or too expensive to use for anything at this point. Being able to compete with a smaller die, GK104, is a huge plus for NV who have achieved a much bigger jump in performance than AMD (per mm^2), the biggest jump since G80 in fact. Added plus is the GK104 memory controller that blows even NV engineers away and makes it possible for the GK104 to function as a high end GPU even in situations that might need high bandwidth. After this AMD launches a price war, bundle war, renaming war and a driver war to make their lineup more lucrative but the damage has already been done and the 600 series has very good rep outside the enthusiast scene.
GTX Titan: GK110 is expensive to produce, supercomputer firms, universities, etc. are screaming for GK110 based teslas which NV provides as soon as they have tens of thousands of GK110s that have low enough leakage and enough working SMXs. K20s are being sold in late september before the launch of the official cards. One of the customers is the Titan supercomputer, needing around 16,000 units.
As usual NV intends to sell the more leaky chips as GeForce ones as has been speculated for well over half a year. Nvidia has a card that can without any trouble beat a 680 and a 7970, the performance difference is much bigger than anything we've seen before, high enough that the card doesn't have any competition at all in the single GPU arena, even less so than the 580. On top of this the card is also almost 100% sure to not have any single GPU competition in the next year or even after that because of AMD and NV being stuck on the 28nm process. Nvidia has also been catching flak for bad compute features lately.
Nvidia doesn't need this expensive chip to compete with AMD and thus has no intentions of releasing it as a normal card part of their normal naming scheme and lineup. That task is reserved for GK114 most likely. However the performance doesn't quite warrant the price of a halo-card-esque pricing of the 690. A decision is made to still release GK110 as Titan for $999 but to load the card with features to make up for the performance that isn't quite up there with the 690. The cooler is similar, users will have full overclocking abilities, etc. And on top of this DP performance is left untouched making the card lucrative for individuals who wish to do GPGPU work but unattractive for big companies. The card is also a huge marketing tool which shows when looking at the emphasis on thermals and OEM systems like those of Origin PC's maingear's etc. The card will enhance NV's brand by setting all the records and being the best. A high price will only reinforce this effect even if it might be bad for the average consumer.
Titan is a product of no competition in the same performance range at all aside from multi GPU setups which don't really matter due to Titan being able to do 4-way SLI, taking less space, being quieter, having the ability to be installed in SFF builds etc.
Pricing has everything to do with the competition. Everything.
So no, your $500 high end GPU isn't going anywhere. If AMD releases the 8970 or the 9970 as powerful enough to make using GK114 or GM104 impossible, NV will be forced to use GK110 or GM100 for normal GPUs. And unless AMD pulls a miracle out of their ass (something like the next version of pitcairn/smaller level chips performing on the level or above NV GK114/GM104) they can't price their cards higher either.
The only thing that allows TItan to be priced at $999 is the huge jump in performance made with kepler. If the manufacturers want to keep raising prices they have to be able to make lower end chips handily outperform previous high end chips. This might have happened once or twice but it is not by any definition the norm and in normal circumstances a price hike of the extreme high end products isn't possible. Due to the huge advances made with kepler NV is most likely going to struggle with Maxwell when trying to maintain similar advances in performance of their chips (they hit the jackpot once, now a new architecture that is better than an incredible one is very hard to do). The jump will most likely be smaller forcing the price of the top of the line chip down. This is of course assuming that AMD keeps putting out GPUs at a steady pace.
Facebook
Twitter