In some of the AMD press conferences, they made the comment that Vega was to compete with Volta. Even for a cut down card, that graph is a little depressing.
Yeah, it is disheartening if this is what is appears to be.
When it comes to improving GPU performance (though the same applies to CPU in a way) is that you can do three basic things, first is increase clocks, second is to increase performance per clock and 3rd is to eliminate internal bottlenecks.
Increasing clocks and increasing performance per clock effectively do the same thing, get more raw performance out of your GPU per second, which in theory improves performance across the board.
The 3rd option will only improve performance in select scenarios, for example, look at AMD's tessellation performance over time. It has increased tremendously, but it only helps in select situations where tessellation is used heavily.
A lot of the changes that AMD has reported for Vega will help to reduce internal bottlenecks, which will help AMD deliver more consistent performance across a wide number of game releases. Even so, it will still take a big increase in clock speeds (or perf per clock) to compete with Pascal on the high-end.
In recent years Nvidia's mantra has been to get more out of less, using compression techniques and other methods that allow them to use smaller memory bus sizes than their AMD competitors. The fact that Nvidia has been able to increase their clock speeds so much also allows them to create high-performance GPUs without increasing GPU core count, which allows them to hit high-performance targets with small die sizes.
Just look at the GTX 980, it was able to deliver GTX 780Ti performance with a much smaller core count and die size. The same can be said for the GTX 1080 replacing the 980Ti, though the GTX 1080 was able to easily beat the 980Ti despite its lower core count. Nvidia then scales these GPUs up to larger die sizes and repeats the process.
Context problems.. The 980Ti at launch was a fair bit faster than the Fury X. In fact, it pretty much decimated it until you got to 4k.
From what I have seen the 1070 is also a little bit faster than the 980Ti. Sure, "If I clock my 980ti to 1500mhz!" etc, but it's surprising just how few of them in reality clock that hard. It's quite easy for a reviewer to do it because most of them were handed golden samples in the first place (why would Nvidia or AMD hand over a stinker?) and so on.
I also think that the 1070 is faster than the 980Ti in DX12. From what I have seen with my Titan X any way.
As I said, who is the performance winner really depends on what you are testing. For the most part the GTX 1070 is the performance leader out of these three GPUs, though again it depends on what you are testing. If you used Dawn of War III the Fury X is the clear winner, but if you use a game like Styx: Shards of Darkness the Fury X is laughably slow by comparison.
To put is simply, the Fury X has a lot of potential that goes underutilised in a lot of games, which a lot of Vega's improvements seem to target. As I have said before Vega will also need a more general performance boost to back this all up, as a better utilised Fury X design will not compete with modern GPUs. What AMD needs is a huge boost in clocks over the Fury X, which is why everyone is looking for clocks of around 1500MHz or higher.
The GTX 1070 does beat the 980Ti in DX12 for sure, especially if the game supports a Nvidia friendly implementation of Async Compute. In most DirectX 11 games, the performance differences are fairly minimal. The 980Ti tends to get higher gains when overclocking, so when both are overclocked they perform very similarly in most DX11 games.
Right now AMD's silence regarding Vega is worrying. Right now we know almost nothing about Vega apart from some of the GPU's core design changes and even then we do not really know how this will change Vega's performance characteristics in games. These changes will really help AMD with the compute market, but only AMD knows how it will affect games.
I really hope that AMD has been influenced by Microsoft's Project Scorpio SoC when creating Vega, as it seems that they have succeeded in getting just that little bit more from AMD's GPU design by making numerous changed based on developer input. These changes could prove very valuable for AMD when it comes to getting the most from DX11/12 games.
