fsync performance results
Thief (239160)
I ran the Benchmark 3 times for each run. Not sure how reviewers do this, it's time consuming and boring =P, anyway... Oh yeah, using Proton 4.11-2
1440p+VeryHigh(custom settings)
fsync & esync Off
Min: 32.63
Max: 93.43
Avg: 57.56
esync On
Min: 54.53
Max: 123.36
Avg: 88.86
fsync On
Min: 55.9
Max: 122.8
Avg: 87.63
Notes: I experienced a couple of app hangs with fsync as reported here
System Specs
Sleeping Dogs Definitive Edition - In-Game Benchmark Tool:
Made sure to run the benchmark first so there would be no shader compilation during testing.
w/Proton 4.11-2
Nothing:
avg: 50.3
max: 88.4
min: 27.7
ESYNC:
avg: 71.9
max: 98.6
min: 48.0
FSYNC:
avg: 74.9
max: 105.1
min: 49.1
Specs: AMD 8350 with 1050ti
GTA V (Social Club version)
Ran benchmark 3 times for each config. fsync isn't turned on by default when using Lutris, used WINEFSYNC=1 to enable.
Lutris + Proton 4.11-2
1440p high(custom)
fsync & esync off:
Min: 44.0735342
Max: 136.070210267
Avg: 99.685273267
esync on:
Min: 46.394250467
Max: 184.974190333
Avg: 130.629943933
fsync on:
Min: 51.347102867
Max: 179.839206
Avg: 127.200167933
Note: I experience one app crash when running a fsync benchmark, not sure how to create debug logs from Wine, so I haven't tried to replicate yet.
Results for Borderlands GOTY Enhanced.
I did crash once when leaving fsync enabled, however I am not sure if that was a one-off or even related to fsync at all.
| Software | Version |
|---|---|
| Nvidia-vulkan | 418.52.18 |
| Proton | 4.11-1 |
| DXVK | 1.3.1-git - https://github.com/doitsujin/dxvk/commit/02d92210adb3369d4c3d7bb06660ca0cde0a2add |
| Kernel | 5.2.5-arch1-1-fsync |
| Game | fsync On / Off | FPS |
|---|---|---|
| Borderlands GOTY Enhanced | Off | 122 fps |
| Borderlands GOTY Enhanced | On | 130 fps |
FPS recorded sitting at main menu, with maximum graphics options at 1080p, plus a custom Reshade config.
GRIP: Combat Racing (396900)
- 1080p Max settings
- Vsync off
- Film Grain & Motion Blur off
- FXAA
Fsync:
Esync:
None:
@Plagman I would like for someone else to try by themselves, but this looks like a pretty huge improvement
Title: Greedfall
Steam ID: 606880
Notes: It required the mfplat patch to boot (Bacuse of the boot video sequences with companies logo).
System:
AMD Ryzen 1700 OCed to 3.6 Ghz
AMD RX 580
16 GB RAM
OS: Fedora 30
Kernel: 5.2.11-201.fc30.x86_64 with f-sync (https://copr.fedorainfracloud.org/coprs/gloriouseggroll/kernel/)
Mesa Version: (Mesa-ACO) https://copr.fedorainfracloud.org/coprs/gloriouseggroll/mesa-aco/
The only difference there was using the PROTON_NO_FSYNC=1 for testing purposes.
It was weird, because the game is not that impressive on the graphics side, the GPU usage was under 30% and the 16 CPU Threads were with a 75% usage. Then I saw a Youtube Video (https://www.youtube.com/watch?v=g_MLhvd24QM) and thought the the main difference could be the f-sync kernel patch, so I got it for Fedora and tried again and the game was running incredibly well. After that I tried disabling the f-sync with the Proton flag to check that was it, and it seems so.
Game: The Outer Worlds
Wine: wine-4.18.r2.ge380c1e9 ( TkG Staging Esync Fsync Vkd3d )
DXVK: 1.4.4-async
Kernel: 5.3.7-17-tkg-pds
CPU: i7-4770
GPU: 980Ti
Logs recorded in-game with exactly the same mouse position & character location, over a 30 second timespan. Recorded with flightlessmango's mangohud DXVK patch
| FPS | Esync | Fsync |
|---|---|---|
| Max | 73 | 70 |
| Avg | 60 | 63 |
| Min | 51 | 60 |
https://flightlessmango.com/logs/13
What's much more apparent in the graphs is that the frametime is so more more stable with Fsync which is nice to see.
Proton versions
proton 4.11-2x3 2019-08proton 4.11x1 2019-08
Launch options
PROTON_NO_FSYNC=1x2 2019-09WINEFSYNC=1x1 2019-08
Upstream links
- ValveSoftware/Proton/issues/2922x2 2019-08
Post your performance results here! If they're a regression, please also report them in the dedicated fsync issue https://github.com/ValveSoftware/Proton/issues/2922.
Make sure you're running the same build of Proton 4.11 on both sides (eg. don't compare 4.2 with 4.11), and the only difference is fsync toggled on or off. (PROTON_NO_FSYNC=1 vs. not)