Is it normal for DX11 to take more than double the time, loading the game world, that DX12 takes?
It took a bit of testing, but I did find a way to reproduce this. I am normally playing on a 100Hz monitor, where load times in DX11 and DX12 are identical - about 10 seconds to get into the game.
Loading time seems tied to frame rate. The game normally enables V-Sync in the menus, which limits them to 100 FPS on my display. If I force V-Sync off, they run at the game's limit of 240 FPS - and load times drop to ~5 seconds.
However, if I limit the frame rate to 24 FPS there is a big difference. DX11 mode takes 30 seconds to get into the game, while DX12 mode only takes 20 seconds.
That's not twice as long, but it makes me think that people seeing slower load times in DX11 are either seeing a similar gap due to using a 60Hz display, or have something set differently with either V-Sync or a frame rate limit.
For example: I use SpecialK to force the game into Flip Mode in DX11 for G-Sync to work properly, and I also use it to limit the frame rate to 90 FPS.
It is not necessary to use SpecialK to get Full-Screen G-Sync working in DX12 mode, so if V-Sync is disabled it's 90 FPS in DX11 vs 240 FPS in DX12 when loading, which halves the load times from ~10s to ~5s. Not because DX12 is faster to load, but because the frame rate is unlocked.
Maybe MFAA does not work on transparent textures, thus 2x is still applied there ?
Or MFAA does not work well in this engine ? But some edges seem to look better on your screenshot.
Also, as the game uses reconstruction, MFAA may not be that helpful.
Before moving to 1080p DLSS in 4K (which looks way sharper and detailed than 1080p with MSAA 4x), I noticed that MSAA mostly had an impact on some edges while moving, on static scenes it did not change that much thanks to reconstruction.
But now I use DLSS, image is sharper, more detailed, and there's no more shimmering on transparent textures, nor aliased edges when moving.
Unfortunately DLSS is only available for RTX GPUs.
4x MFAA is 4x MSAA plus jittered sample positions and temporal accumulation. Enabling MFAA does not halve the sample count; it's supposed to "double" the effective sample count after temporal accumulation - so 4x MFAA ≈ 8x MSAA, according to NVIDIA.
It clearly should not be used with
Control, but as I said before, I have yet to find a game where MFAA actually looks better. I'm not convinced it actually works as intended.
