I strongly believe this is the case. There is a healthy number of users that upgrade frequently, I'd say perhaps every other year. They aren't buying top of the line GPUs, but they go from GTX 670 to GTX 970 and will likely go to GTX 1070 because each upgrade was roughly two years apart and came with a good performance increase.
Contrary to this belief on AT that a lot of gamers upgrade their GPUs annually or even every 2 years, it's simply not true for the vast majority of PC gamers. NV's own Investor Day 2016 data shows that 70% of dGPU owners are using Kepler and older architectures.
http://forums.anandtech.com/showpost.php?p=38190615&postcount=2397
What's critical here is that according to AMD, less than 10 million PC gamers have GPUs as powerful as an R9 290X/970 and according to NV,
70% of their userbase are on Kepler and below.
That actually means the vast majority Fermi and Kepler owners have suffered great performance degradation against AMD, while also delivering worse price/performance (since similarly or higher priced Fermi and Kepler cards are easily getting demolished by back then competing AMD GCN HD7000/R9 200 series).
There are other points that should not be ignored. A more future-proof architecture allows budget and mainstream gamers to delay their upgrades or buy a lower tier card that ends up performing just as well as a flagship years later (i.e., $300-350 7950 vs. $450-500 680, $300 280X vs. $450 770 4GB, $400 R9 290 vs. $700 GTX780Ti, etc.)
Furthermore, when it comes time to resell the old NV card or pass it on to a family member/friend, that 2nd or 3rd user is yet again suffering by getting a far worse graphics card than AMD's competing product. What exacerbates the situation is most NV cards retain their resale value longer. In the used market, it's easy to come across a situation where an AMD card 1, 2 or even 3 tiers above sells for barely more than an NV card. That means not only do modern NV cards cost more to start, age worse, make less or no $ with crypto-currency, but even in the used market they still cost more for 2nd hand buyers. Awful. This only highlights that A LOT of the value for NV cards is
perceived value, not measurable or quantifiable one.
The more lucrative buyer are going from GTX 680 to 780 to perhaps 780 Ti to 980 and probably 980 Ti.
This I am more inclined to believe. Gamers who like to stay on the cutting edge and buy $500-550+ high-end cards are far more likely to upgrade every 12-24 months. It also makes sense to roll-over the resale value and reinvest in into a next gen card before your existing card's value plummets (i.e., $549 R9 290X or $699 780Ti -> $330 970 suggests it's better to dump all flagship cards prior to the launch of new cards. This applies to AMD too).
I think AMD missed the bus with their uarch and their release schedule. Yes, it's a win/win for consumers, but that is slowly what is killing AMD in marketshare and overall profits.
That's
not what's killing AMD's market share. Something else is.
AMD went from selling 7.7M dGPUs per quarter in Q1 2011 to just 3.55M by
Q3 2014.
Post #174 here:
https://www.techpowerup.com/forums/threads/nvidia-geforce-titan-x-12-gb.210662/page-7
That means
before GTX970/980 launched, AMD already suffered catastrophic market share losses. This proves that it was NOT GTX970/980 that dealt the greatest market share loss to AMD.
Look at the AMD Enthusiast:
Bought HD 7970, OC it skip 7970 Ghz, bought 290X, OC it skip 390X, and possibly buy Fury X. The jumps between flagships was minimal in the same time frame that NV release their own flagships.
Except if we use Steam Hardware Survey that NV fans love throwing around, it also shows as the graph above that AMD suffered the greatest losses in the sub-$330 dGPU market segments over the last 5 years.
AMD Radeon HD 7900 Series (7950, 7950 V2, 7970, 7970Ghz) =
1.51%
vs.
NVIDIA GeForce GTX 670 0.49%
NVIDIA GeForce GTX 660 Ti 0.44%
NVIDIA GeForce GTX 680 0.34%
Total for GTX660Ti-680 series =
1.27%
That means HD 7900 series easily outsold their NV competitors using Steam.
Moving on to the next gen
AMD Radeon R9 200 Series =
1.57%
vs.
NVIDIA GeForce GTX 770 1.23%
NVIDIA GeForce GTX 780 0.53%
NVIDIA GeForce GTX 780 Ti 0.26%
Total for GTX770-780Ti =
2.02%
NV did better but not enough to command 80% market share.
That means clearly, where AMD is bombing is the low-end and mainstream market segments.
Hint*
NVIDIA GeForce GTX 960 3.14%
NVIDIA GeForce GTX 750 Ti 2.71%
NVIDIA GeForce GTX 760 2.07%
NVIDIA GeForce GT 620M 1.95%
NVIDIA GeForce GTX 860M 1.79%
NVIDIA GeForce GTX 660 1.59%
NVIDIA GeForce GTX 650 1.51%
NVIDIA GeForce GT 630 1.44%
NVIDIA GeForce GTX 750 1.29% <=> This turd is more popular than the
entire GTX660Ti-680 series per Steam
NVIDIA GeForce GT 730 1.14%
NVIDIA GeForce 840M 1.06%
NVIDIA GeForce 210 1.01%
etc.
http://store.steampowered.com/hwsurvey
You should get the point by now. Other than GTX970, AMD is getting hammered in the $249 and below desktop and mobile dGPU market segments. That is exactly where NV has the worst products and where AMD can capitalize the most with Polaris 10/11. Target $129, $149, $179, $199, $229 and $249 prices.
On the two year cycle, you had 7870 having to hold out until 280X/285/380X which was a less upgrade versus some one going from 670 to 970, and if that buyer was wise, they jumped on bargain sale 290/290Xs and will probably stay there skipping 390/390X and possibly Polaris 10.
The comparison you made is strange. You compared jumping from 7870 to 280X as similar to 670 to 970 but in the AMD line-up, an equivalent NV upgrade path you outlined would be HD7950 -> R9 290.
AMD is as much fighting NV as they are fighting themselves.
Can't be an explanation because ATI 9000 series was more future-proof than GeForce 4 and 5, and ATI X1800/1900 series was also more future-proof than GeForce 7 and yet in those days ATI had a lot more market share.
A much easier explanation is that OEMs and pre-builts preferred NV's products because they could use 250-400W garbage PSUs and combined that with any NV card in the sub-$300 space. It's pretty simple - NV is perceived as the more valuable brand and uses less power. OEMs don't even have to choose in this case. For AMD to win OEMs they have to beat NV or offer prices so low, the OEMs will take it.
Another dynamic is happening in PC gaming. The newer generation PC gamers do not know any of the history of ATI vs. NV. As a result, anyone who is young building a new system is following Twitch, Twitter, YouTube and seeing NV everywhere. Similarly, they only think of AMD cards as hot and loud because they don't know the history of ATI's graphics prior to the year 2011-2012. This is also why AMD has a massive perception problem nowadays. In many ways, newcomers to PC gamers or console to PC gamer converts don't know much and the safest brand is NV. This is also why so many NV users outside of technical forums such as ours are uniformed because they are inexperienced PC builders/just newcomers to PC DIY space in the first place. NV has been able to better capitalize on this new demographic of 18-27 year old gamers.
This is why it's critical that AMD targets sub-$299 market segments first even if it means conceding the $350-650 segments for 6 months.
EDIT: I'd also like to add, that AMD's DX12 advantage hasn't really paid off just yet. Pascal could possibly change that.
Of course it hasn't paid off since we don't have many DX12 games, and the ones we have are not big sellers. By the time more DX12 gamers arrive, the 70% of NV users who are still on pre-Maxwell cards will start upgrading to Pascal in record waves. It's in AMD's best interest to capture them with Polaris 10/11 before GP106/107 have a chance to land. Since the vast majority of PC gamers don't spend above $350 on a dGPU, this is a key opportunity for AMD. As far as Fermi, Kepler and sub-980Ti Maxwell owners go, NV will be more than happy if Pascal performs the best in DX12 while all those older NV architecture fall apart. Since most gamers are too brand attached, even if they own GTX500/600/700 series, instead of not giving NV $$ after being burned so hard, they'll just brush it aside and line-up to give NV their monies again. After all, we know if AMD pulled off stunts such as blocking mobile dGPU overclocking only later to call it a bug, released 3-4 drivers that killed its GPUs, VRAM gimped a card only to have no remorse thereafter, there is NO way there would be no consequences. Yet, even when NV messes up greatly, people keep buying. Just goes to show the power of mind-share, perceived value and branding.
Soon more DX12 games should follow:
Watch Dogs 2
http://www.kitguru.net/gaming/matthew-wilson/watch-dogs-2-to-feature-dx12-and-amd-optimization/
Total War Warhammer
http://www.gamersnexus.net/news-pc/2349-amd-and-total-warhammer-enter-dx12-partnership
Deus Ex Mankind Divided
http://wccftech.com/deus-mankind-divided-revealed-trailer-features-dx12-tressfx-support-dawn-engine/
If AMD manages to score wins in some of these upcoming DX12 titles, it'll will probably be too late anyway since by then NV will have replaced 970/980/980Ti with cards that are much faster and/or cheaper; thus saving their face.