But otherwise, for anyone purchasing a card.... this is a nonissue. Purchase whatever card plays the games you want now, and realize that future games may benefit a different vendor, but the competitors card in the same price bracket will still be within 10% or so of your card.
1) But the differences under DX12 thus far are more than 10%.
In Ashes, 390 beats 970 by
16% at 1080P high preset.
In Fable, 290X beats 970 by 13%.
So 13-16% advantage for 390/290X that compete against 970. 16% difference is nearly as much as between a $330 GTX970 and $550 GTX980 at launch.
2) You aren't considering the entire context.
For example, at 1080P, R9 380 is about
8% faster than the GTX960 4GB. If DX12 games give GCN 13-16% benefit on average, that 8% turns into 1.08 * (1.13-1.16) = 22% to 25% advantage! That's HUGE.
3) If you only compare cards in similar price brackets, you miss the other aspect - price/performance.
For example, a $450 GTX980 is 12% faster at 1080P and 8% at 1440P compared to a $280-300 R9 390. All of a sudden 13-16% boost under DX12 games for the 390 means that someone who bought a 980 just threw $150-170 into the toilet.
Similarly, the 970 becomes very difficult to recommend against the 390 since 13-16% benefit under DX12 ensures 390 wins in raw GPU performance and VRAM. That's a huge key battleground in Q4 2015/Q1 2016 since 970/980 and 390 will still be sold for the next 3-5 months before 16nm GPUs launch.
And of course we have R9 280/285/280X/290 vs. 950/960 2GB/960 4GB cards. 13-16% more performance for the GCN makes it literally impossible to recommend the 950/960 series for gaming for anyone who intends to keep their cards for 2-3 years and will play DX12 games. Do we know for sure that the 13-16% advantage is an accurate estimate, no, but it's a trend that's emerging in 2/2 DX12 games.
4) You are not taking into account how much DX12.1 was touted as a key selling feature for Maxwell over GCN in marketing documents and online. It was plastered all over by the media/NV. How many gamers were on the fence between AMD/NV cards and picked Maxwell since they thought over the next 2-3 years of ownership it was safer to go with a 'better' DX12-compatible + newer Maxwell architecture? Don't you think this type of marketing is misleading to the consumer?
5) Also, while you can make the argument that DX12 performance for modern cards doesn't matter because for next generation DX12 games gamers will need to upgrade to future GPUs, you would be mostly right. However, since not everyone times the market and some gamers who will buy GPUs over the next 3-5 months have extra information to make a more informed decision. What if the DX12 benchmarks we've seen do not even use AC extensively and thus the GCN architecture isn't even showing the full potential? We don't know that either but looking at GTX470/580 vs. HD5850/5870 and seeing how lack of good tessellation leveled HD5850/5870 cards, I'd rather recommend the more safer architecture that supports more next gen features and performs better with them. Will 13-16% be enough to play 2016-2018 DX12 games? No but it's a
free bonus on GCN at least from the two benchmarks we have.
6) Fable Legends is a UE4-based titles. From the DX11 UE4 benchmarks we have, Maxwell walked all over GCN. The fact that a 925mhz 7970 pummels the 680, 285 destroys the 960 and 290X beats 970 under UE4 is an eye-opener. This could mean that under a more brand-agnostic DX12 game engine, GCN would show even greater performance advantages. Given how well UE4 games have generally run on Maxwell cards, and possible lack of wide use of AC (on a % basis from the total game engine code) in Fable Legends, things could be even worse for Maxwell than it seems. Counter to that point is that NV could strengthen its efforts on GW front even more and influence the use of AC in DX12 games to diminish any advantage GCN may have. We'll have to see what happens over the next 6-12 months with DX12 games to try and get a better picture.
So while the current recommendation for $650 980Ti vs. $650 FuryX doesn't change, the rest of NV's line-up looks far worse now than it did prior to the DX12 scores based on what we have seen thus far. Luckily for NV, if Maxwell really takes a huge performance hit with AC+graphics context switching under DX12, it's already October 2015 which means they only have to coast for ~6 months before they can close the chapter on Maxwell as the hype for Pascal will take over.