It is the CPU, but it's because it's good, the problem is in reverse guys.
The FX 8350 should perform extremely poorly compared to a 2500k under the right circumstances. The more that CPU is taxed, the more poorly it performs on the logical core level, down to about 50% of what it should be.
If the benchmark is taxing both logical cores, then DX11 is getting a little more than half a physical core, and a single 2500k core is more powerful than a single 8350 core to start with. ASADDF, that's why your DX11 numbers are so poor, you can't feed even a 770 with a partial core running off already loaded shared assets. Under DX12, it spreads that utilization across the four sets more evenly, and you're GPU bound without even trying.
The 2500k is probably enough juice to feed a 770 GTX with just one core. Powerful individual cores mixed with lightweight GPU's aren't going to get much out of DX12, they're already making the most of their assets to begin with. Having a 2500k myself, with a 680 GTX, very close in processing power, I overclocked to 4.3Ghz and saw only the slightest of improvements in DX11, I was only just barely cpu bound at stock clocks.
There does seem to be something wrong here though. Eviator should have much better scores, going by yours. Do you perhaps have a 4GB 770 GTX, and Eviator has a 2GB version? It would take much more work to run off 2GB of ram. If you guys actually have the same card, something is majorly hosed on his system, massive performance problem somewhere.
