FFXIV runs poorly in dx11 mode compared to dx9

[ashkitten]

in dx11 mode the graphics settings seem to barely affect framerate at all (only about a 10 fps difference between minimum and maximum), and I get about 70fps or so in the lavender beds, dropping to sub-50 in crowded places like limsa lominsa aetheryte. in dx9 mode, it maintains a fairly stable 110fps in the lavender beds on maximum settings, getting up to 210fps when turned down to minimum.

Software information

maximum settings, wine fsync enabled

System information

• GPU: AMD RX 5700
• Driver: AMDGPU (Linux 5.8.10-zen) / Mesa 20.1.7 / RADV (ACO)
• Wine version: lutris-5.7.10
• DXVK version: v1.7.1-2-g743f309

Log files

https://gist.github.com/ashkitten/ddbf3097df96f2391c0d1fac63404807

[doitsujin]

D3D11 is consistently faster than D3D9 mode on my end when using comparable graphics settings, not really sure what your problem is. Do note that D3D9 more does not support stuff like HBAO+ and tessellation, which can cause a fairly significant performance hit.

Dropping below 50 FPS in crowded Limsa is normal and happens on Windows as well, unless you're running something like an overclocked 8700K or better.

Also, please make double-sure that fsync actually works on your end (or just attach a full wine log).

[ashkitten]

minimum graphics settings should be comparable though, right? i only get around 80-90fps in the lavender beds on dx11 while dx9 gets over 200. i've seen the fsync: up and running. message, so unless there's another way to check if it is working i don't know...

[doitsujin]

And are you CPU-bound or GPU-bound?

I have absolutely no info to work with right now, other than "works fine for me" (both on my RX 480 and mobile RTX 2060).

[doitsujin]

You could also try running the old Heavensward benchmark with both D3D9 (max settings) and D3D11 ("Maximum (DirectX 9)") and compare; if there's a significant difference in scores then there might be a driver issue, if not, then there simply is no problem.

[ashkitten]

seems like they get very similar scores on that benchmark. i realized that while i have fsync working in the lutris build of wine (which i had to use to get dx9 working) it doesn't log anything about fsync in the wine-staging system build, so it's possible that might be the issue for me (along with dx9 not supporting the same settings). sorry for the noise, then.

[SkyLeite]

I'm experiencing the same issues on similar hardware (RX 5700 XT + Ryzen 5 3600). Esync is enable and confirmed running on lutris logs.

Wine build: lutris-5.7-10-x86_64

DirectX 11 Maximum settings: ~100FPS
DirectX 11 Minimum settings (Standard Laptop): ~105FPS

DirectX 9 Maximum settings: ~145FPS
DirectX 9 Minimum settings (Standard Laptop): ~200FPS

I tried setting the maximum tesselation level to 8 (as AMD cards are pretty bad with tesselation IIRC) but it didn't seem to help. I also enabled the DXVK HUD to take a look at what's going on, and it seems like with DX11 Minimum Settings, my GPU usage never goes past ~65%. Does this help in any way?

https://cdn.discordapp.com/attachments/201788332743720960/768233827347726396/unknown.png

[doitsujin]

Again, not a DXVK issue. I don't know what's broken on your end but something clearly is, maybe try without Lutris and double-check the benchmark I linked above.