This set of benchmarks uses only the frame times and averages from the “heavy” third of the benchmark scenes. In theory, this should put more emphasis on the DX12 and CPU implementation for each combination of hardware.
With an emphasis on the more demanding set of tests, the NVIDIA results change a little bit over the previous page. At 1080p / Low the DX12 path is about 27% faster than DX11 though bumping that to the High quality preset (without AA) again makes DX11 slightly faster than DX12. The AMD R9 390X results again show massive improvements from DX11 to DX12 (as much as 2.28x!) though we also find the gap between the DX11 GTX 980 results and DX11 R9 390X results has grown too. At 1080p / Low the GTX 980 is nearly 90% faster than the R9 390X using the DX11 code path. But once we enable DX12 both cards are on even ground.
Again, the 6700K results are nearly the same as the 5960X with just a slightly global step back in total frame rate. NVIDIA only has a single result that scales from DX11 to DX12 in a positive manner while AMD’s lower DX11 results show massive improvement.
And again, the Core i3-4330 is where we start to see NVIDIA’s GeForce GTX 980 DX11/DX12 performance start to perform as we expected with slightly but repeated gains moving to the new API. AMD’s gains are much larger (in relation to the same GPUs DX11 scores) and the net result is even DX12 performance in Ashes of the Singularity.
Wash, rinse, repeat. Though here the AMD Radeon R9 390X is faster with DX12 enabled across the board.
Another cases of the same-sies; NVIDIA gains less from DX11 to DX12 but does that with a much higher starting point of DX11 baseline performance.
Our per-GPU scaling graph while focusing only on the heavy batches performance shows nearly the same pattern as we saw in our average results on the previous page. The NVIDIA GPU and driver just don’t have area to improve with our two fastest processors but see consistent, but small, advantages with the slower three processors tested. It is worth noting that these gains are larger from DX11 to DX12 with the GTX 980 as we push the emphasis to the most difficult third of game scenes.
The gains grow for AMD here moving from lower DX11 scores to much faster results with the DX12. Whether it be because of poor optimization at the DX11 level or better optimization at DX12, it seems obvious that AMD gamers are going to see some big benefits with the DX12 code path in Ashes of the Singularity.
Finally, this graph attempts to show which card has an advantage on each processor with each setting and with each API. We are using the R9 390X as the baseline score, meaning that any positive percentage result means the GTX 980 is faster while a negative score means the R9 390X is faster.
Wow, those DX11 gains got even bigger for NVIDIA’s GeForce GTX 980 when we focus only on the heavy batch scenes in the benchmark! We are seeing nearly 2x the performance with the Core i7-5960X and Core i7-6700K in some instances. And in general, the total DX12 performance between the R9 390X and the GTX 980 is a little bit closer on the heavy data set than it was on the total, average results. Because the AMD hardware had more ground to make up to overtake the wider DX11 gap, this makes sense.