It was a current GPU at the time of the 1070's release. That's why reviews of the time do that comparison. They ended up having to halve their price to shift them. That's why the 1070 got the high review score and that's why people bought them. I can't see which alternative AMD GPU you think was better value:
- The Vega 56 came out a year later, was marginally better but cost 50% more.
- The Vega 64 also came out a year later, was 25% better but cost 100% more.
- The 5700 XT was 20% better but cost 100% more and came out 3 years later.
- The RX 590 came out 2 years later and was the same price but offered 25% less performance.
- The R9 Fury X came out the year before and offered less performance for a far higher price.
Fury X was 28nm, GTX1070 was 16nm. End of comparison. Absolutely o point comparing those, as 28nm was 2011 tech and 16nm was 2016 tech. Much more valid comparison is RX480 vs GTX1070, and RX480 offers better FPS/price ratio. That reason alone is enough not to make GTX1070 not 100/100 card.
Review score was given when review was made, so anything that happens after that doesn't matter at all. When giving review score, there was no real knowledge what comes on future so...
I got most of the data from UBM as it was easier to collate. The really telling thing they show on UBM is market share - roughly 10 x the number of people bought the 1070 in comparison to the other (roughly) equivalent AMD cards. I'm guessing more people agreed with that 10/10 score than you think.
We are talking about review score here, not about how many people bought what. Again, score was 100/100, not 10/10, and you are entirely missing the point.
GTX1000-series was basically nothing more than die shrink (some small improvements) of GTX900-series. So how just die shrink justify 100/100 score? If there would have been die shrink WITH architecture improvements AND huge amount of new features (like still quite useless Ray Tracing acceleration but that is still good new feature), then what score would have been? 200/100?? Problem is that when giving 100/100, it indicates nothing could have been done better. With GTX1070, there was very much that Nvidia could have done better.
Another thing is that Nvidia got rewarded for crappy work (skipping 20nm entirely). So 1000-series was much better than 900-series. Using this same logic. Let's assume that AMD launches Zen4 (5nm) and Zen5 (3nm) on some future. Now, AMD would get better score for Zen5 if they just skip Zen4 entirely. That way boost compared to predecessor is larger
So overall, you have some points but they are not so valid when staying on this scoring system.