DroidDeck · AYANEO Pocket FIT · Adreno 750 · 144 Hz

DD 0.6.0 Stable and now

DroidDeck's performance fixes stop DroidDeck, gamescope and Wine from waiting on each other every frame, and with PR #1697 Windows games get the fastest core as CPU 0. Since PRs #1706–#1708 gamescope and the compositor handle each new frame as soon as it arrives, and Steam's start no longer slows the first seconds of a game. The first fixes started as the Performance mode switch; since PR #1650 every session runs that way, and Steam's Frame Limit in the Quick Access menu is the only frame limit.

×5.0Vulkan frame rate

×17Direct3D 12 frame rate

×5.8Direct3D 11 frame rate

×1.9GTA IV benchmark

Every API: DD 0.6.0 Stable, now, and a Winlator fork

Vulkan

×5.080 % of the Winlator fork

456

2,280

2,846

OpenGL

×1493 % of the Winlator fork

205

2,802

2,999

Direct3D 12

×1794 % of the Winlator fork

282

4,935

5,233

Direct3D 11

×5.888 % of the Winlator fork

666

3,895

4,408

Direct3D 10

×3.689 % of the Winlator fork

1,088

3,934

4,417

Direct3D 9

×6.292 % of the Winlator fork

673

4,176

4,561

Direct3D 8

×1258 % of the Winlator fork

225

2,663

4,611

DirectDraw / D3D7

×3264 % of the Winlator fork

72.8

2,345

3,678

AIO Graphics Test 2.2.0, average fps, no frame limit, Turnip, same Pocket FIT; bars scaled per API; the percentage is DroidDeck now as a share of the Winlator fork. DD 0.6.0 Stable: the 2026-10-22 release had none of these fixes; measured 2026-10-26 with every fix switched off (the old Performance mode switch off), 15 s runs, the middle of three: frame rates swung widely then (the range is under each screenshot below). DroidDeck latest: 2026-10-28, DroidDeck main 213c2da5 (every fix on, including PRs #1706–#1708); DirectDraw / D3D7 on the PR #1715 build 8c94cb85, merged as main 9d82df69 (the review changes after it do not touch the default path), DroidDeck-Proton 11-11.9, on battery, 30 s runs, best of two. Winlator fork: 2026-10-21, 15 s runs, best of two. DirectDraw / Direct3D 7 run through WineD3D on OpenGL; since PR #1715 OpenGL no longer waits for each screen refresh while Steam's Frame Limit is off (AIO D3D7 144 → 2,345 fps), and the Frame Limit still holds it when on (72 → 71.9 fps). Direct3D 8 runs through WineD3D by default; the per-game PROTON_DXVK_D3D8 option raises AIO D3D8 2,408 → 3,097 (measured 2026-10-25).

Every API: best of two runs and its 1 % low

API	Average best of two	1 % low same run	Before 2026-10-27 · PR #1697 build	Change average · 1 % low
Vulkan	2,280	887	2,136 · 824	+7 % · +8 %
OpenGL	2,802	820	2,762 · 818	+1 % · 0 %
Direct3D 12	4,935	1,060	4,806 · 1,058	+3 % · 0 %
Direct3D 11	3,895	787	3,539 · 663	+10 % · +19 %
Direct3D 10	3,934	783	3,835 · 766	+3 % · +2 %
Direct3D 9	4,176	1,005	4,019 · 896	+4 % · +12 %
Direct3D 8	2,663	797	2,546 · 701	+5 % · +14 %
DirectDraw / D3D7	2,345	620	144 · 104	×16 · ×6.0

AIO Graphics Test 2.2.0, 30 s runs, no frame limit, Turnip, on battery with the fan on full; each API ran twice and the run with the higher average is shown, with that run's 1 % low (the frame rate the slowest 1 % of frames stayed above). Before: the same test on 2026-10-27 · PR #1697 build. AIO varies by a few percent between identical runs.

Games: DD 0.6.0 Stable and now latest runs 2026-10-27, PR #1697 build

DiRT Showdown

×1.2

92.2

108.4

Before: DD 0.6.0 Stable had none of these fixes; measured 2026-10-25 with every fix off, average of all runs. Latest: 2 runs (108.3 · 108.4).

GTA IV

×1.9

32.4

61.4

Before: 0.6.0 Stable itself not measured; this is the latest build with only the fastest-core fix off (GTA IV pins its main thread to CPU 0). Latest: 2 runs (61.1 · 61.7).

Average fps; each test's bars are scaled to that test. DiRT Showdown's and GTA IV's built-in benchmarks. Latest: DroidDeck built from PR #1697 (bec7a61f), every fix on, DroidDeck-Proton 11-11.9, on battery, Turnip; one of the two DiRT runs was on the PR's first build (41f1a7aa), the same code for DiRT.

DroidDeck and a Winlator fork, same games 2026-10-27 to 29

DiRT Showdown benchmark

GTA IV graphics settings

GTA IV benchmark

Dead Cells main menu

Dead Cells gameplay

Another Crab's Treasure graphics settings

Another Crab's Treasure gameplay, beach

Another Crab's Treasure gameplay, tide pool

Screenshots taken on the same Pocket FIT. DiRT and GTA IV on 2026-10-27, same in-game settings. DroidDeck: DroidDeck-Proton 11-11.9; the DiRT run is from before the fastest-core fix, which doesn't change DiRT. Winlator fork: its Proton 11 ARM64EC container. The overlay in the corner tells them apart: green is DroidDeck's, pink is the Winlator fork's. Dead Cells on 2026-10-28, one moment each, not an average. DroidDeck: the PR #1715 build 8c94cb85 (the code now in main 9d82df69), DroidDeck-Proton 11-11.9. There the game's own frame rate is in Steam's bar across the top; DroidDeck's green box counts the frames sent to the 144 Hz screen. Another Crab's Treasure on 2026-10-29, one moment each, not an average, same in-game settings (1280 × 720, High, textures Highest, V-Sync off, no FPS limit), frame generation off in both. Its numbers are read off each app's HUD: frame rate, frame time, GPU and CPU load, battery power draw. DroidDeck: the PR #1763 build a13c038c, DroidDeck-Proton 11-11.9. There DroidDeck's overlay is the new blue HUD at the top; the white panel on the left or along the top is the Pocket FIT's own overlay.

GTA IV benchmark 2026-10-27

GTA IV's built-in benchmark, 1280 × 720, High / Very High, VSync off, with the fastest core as Windows CPU 0 (PR #1697).

DiRT Showdown: DD 0.6.0 Stable and now

DD 0.6.0 Stable · fixes off, 2026-10-25

Run 1: 92.2 avg, min 61.6
Run 2: 93.3 avg, min 62.3

Average of all runs: 92.2 fps

Latest · 2026-10-27

Run 1: 108.3 avg, min 80.9
Run 2: 108.4 avg, min 81.2

Average of all runs: 108.4 fps

DiRT's built-in benchmark (Nevada, 8 cars), average fps from DiRT's results file. Before: the runs above plus 4 follow-up runs on 2026-10-25 with every fix off. Latest: runs on the PR #1697 build.

Battery and the frame limit measured 2026-10-26

No limit

DD 0.6.0 Stable

10.0 W

Game: 826 fps
Busy cores: 2.6 of 8
GPU busy: 17 %

3 runs, 9.0–10.8 W

With fixes

14.1 W

Game: 2,990 fps
Busy cores: 5.0 of 8
GPU busy: 29 %

3 runs, 13.3–15.0 W

60 fps limit

DD 0.6.0 Stable

9.4 W

Game: 591–1,622 fps
Busy cores: 2.6 of 8
GPU busy: 11 %

3 runs, 8.4–11.0 W, only the screen is paced

With fixes

7.1 W

Game: 60.0 fps
Busy cores: 1.3 of 8
GPU busy: 5 %

3 runs, 7.0–7.2 W, the game is held

Whole-device power from the battery (unplugged, screen brightness fixed, fan on high), averaged over each run's benchmark window; busy cores = CPU use across all 8 cores; AIO Direct3D 11 cube, 45 s runs, Turnip, measured 2026-10-26 on the Performance mode test build. With no limit the fixes draw 3.6× the frames for 42 % more power. The 60 fps rows used the app's own frame limiter, which DroidDeck has since replaced with Steam's Frame Limit (PR #1616, PR #1650); they have not been re-measured with Steam's limit.