Just to clarify, I tested with an i7 2600K clocked at 4.8 GHz, a decently high OC yet still attainable with conventional air cooling provided you have a bit of luck. To emulate the i5 2500K performance, I simply disabled HT. The overclock stayed, as I figure if you have a K edition processor, you are going to overclock it.
The results more or less speak for themselves, but I'll lay it out.
On release, the GTX 670 SC+ performed ~20% better than the HD 7870 Myst. The Myst is a cutdown HD 7950, but even in AMD branded games like Deus Ex: Human Revolution, the GTX 670 SC+ is still around 15% faster. They are in different performance brackets and the launch price reflected that too (270 USD vs 400 USD). So I hope no one is delusional to think that the HD 7870 will beat the GTX 670 outright.
Rise of the Tomb Raider with DX 12, and on the High Preset.
GTA V with DX 11, and on the High Preset.
Crysis 3 with DX 11, and on the High Preset.
All at 1080p.
Emulated i5 2500K results:
Based on the average FPS, the GTX 670 is ~10% faster than the HD 7870 Myst.
i7 2600K results:
Based on the average FPS, the GTX 670 is ~13% faster than the HD 7870 Myst. Compared to release, AMD has gained 5-10% on Nvidia depending on the game.
So why did the GTX 670 perform so much better with the i7 test and the HD 7870 perform better in the i5 test?
It's fairly simple. AMD's problems lay with single threaded command list drivers. This means that to mitigate the problem, a CPU with decent IPC is needed. In essence, overclocking the CPU was able to remove any significant bottlenecks for the AMD card. To add, part of the point of DX 12 is to reduce the CPU overhead, removing potential for the CPU to be a bottleneck. This is why the AMD card did not lose ground in ROTTR when moving up to the i7 test, because the GPU was already the main bottleneck for the game.
Now, with Crysis 3 and GTA V, the reason that they saw improvements with the i7 2600K on the GTX 670 is because they are CPU intensive games. To add, it was tested on High Settings. Meaning that both games are more CPU bound than if they were tested at Ultra Settings. Both are games that still boast performance improvements when moving above 4 threads. Once the thread count was increased, the GPUs became the primary bottlenecks, and the GTX 670 benefitted more from that. The HD 7870 gained frames too, but the difference was marginal compared to the GTX 670's noticeable bump.