Games and 3D benchmarks, fullscreen and without vsync, in Omarchy's session on every machine of the lab, under Hypoland and under stock Hyprland 0.57.1. Every cell is Hypoland / Hyprland: green where Hypoland is at least 5% better, red where it is at least 5% worse; a dash where the benchmark does not run or there is nothing to count. The game's frame rate is what the game drew; the frames on the screen are the compositor's page flips a second, which stop at the screen's refresh rate (120 Hz on the Galaxy Book).

• Xonotic, Low
• Xonotic, Ultimate
• GravityMark, OpenGL
• GravityMark, Vulkan
• SuperTuxKart, Vulkan
• Yamagi Quake II, OpenGL 3
• Yamagi Quake II, Vulkan
• vkmark
• ElectroBench, guns
• ElectroBench, ocean
• ElectroBench, pool
• ElectroBench, lines

Xonotic, GravityMark and ElectroBench are from the night of the lab page, alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2; SuperTuxKart, Yamagi Quake II and vkmark came to the benchmark after it and are from a short run, 2 fresh sessions of each compositor per machine (Hypoland 3 on some), in the afternoon of 2026-11-11, Hypoland r964 (the compositor of r957 with two small fixes) against Hyprland 0.57.1-2: few sessions, so read those tables as a first look. A fullscreen game under Hypoland can go straight to the screen (direct scanout): Hypoland measures each fullscreen window both ways and keeps the better. The Mac mini 2014 had no monitor for either run (it has a headless output in the short run, so no frames on the screen); the Mac mini M1's display driver reports no page flips. On the X200 Hyprland runs with Omarchy's GLES 3 override (omarchy#14764), which the games get too. Raw data, every session's values and the scripts: the night, the short run.

Xonotic 0.8.6 plays its benchmark demo the-big-keybench fullscreen without vsync at the Low effects preset, as Phoronix's test does: a native Wayland client through SDL 2. The demo is stopped after the measurement; the game's frame rate is what it drew until then.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

The same demo at the Ultimate preset. Not on the AMD machine: its r600 GPU locked up under the presets above Low.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

Tellusim's GravityMark 1.89 fullscreen without vsync, an X11 client through Xwayland, its own mean frame rate over its 167 s run. Its OpenGL renderer needs OpenGL 4.5 and crashes on several machines, so few rows.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

The same with its Vulkan renderer: not on the AMD machine (no Vulkan) nor the X200.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

SuperTuxKart 1.5's --benchmark: the replay of a race on black_forest, Vulkan, Ultimate effects, no frame limit and no vsync, fullscreen at the monitor's size. The game's frame rate is the race's frames over its time; its Steady FPS, one of Phoronix's two charts, is a rate most of the race's frames reach, so a few long stalls make it 0.

Medians, which over two sessions are the lower value: 2 fresh sessions of each compositor per machine (Hypoland 3 on some), in the afternoon of 2026-11-11, Hypoland r964 (the compositor of r957 with two small fixes) against Hyprland 0.57.1-2. The stalls come with Hypoland's measured direct scanout (render:direct_scanout = auto, the default) and are being looked into; on the Mac mini M1 the game cannot make its swapchain under it and gives no result, with direct scanout off it runs.

Yamagi Quake II 8.10 plays the timedemo of q2demo1.dm2 over and over, fullscreen, with 16x anisotropic filtering and 16x multisampling (8x on Intel), as Phoronix's test does.

Medians, which over two sessions are the lower value: 2 fresh sessions of each compositor per machine (Hypoland 3 on some), in the afternoon of 2026-11-11, Hypoland r964 (the compositor of r957 with two small fixes) against Hyprland 0.57.1-2.

The same with its Vulkan renderer (ref_vk 1.0.2; 4x multisampling on the Adreno).

Medians, which over two sessions are the lower value: 2 fresh sessions of each compositor per machine (Hypoland 3 on some), in the afternoon of 2026-11-11, Hypoland r964 (the compositor of r957 with two small fixes) against Hyprland 0.57.1-2. On the Mac mini M1 the game outruns the display on the plane under Hypoland: more frames drawn.

vkmark 2025.01 runs its 13 scenes of 10 s each fullscreen in mailbox mode: it draws as fast as it can and the screen shows the newest frame. The score is its own.

Medians, which over two sessions are the lower value: 2 fresh sessions of each compositor per machine (Hypoland 3 on some), in the afternoon of 2026-11-11, Hypoland r964 (the compositor of r957 with two small fixes) against Hyprland 0.57.1-2.

ElectroBench's first scene, OpenGL 2.1, so the X200 too: 110 guns drawn face by face with glBegin, 60 s without vsync, fullscreen and in a window, a native Wayland client. The frame rate is the scene's own mean.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

Scene 2, OpenGL 3.3 core, 45 s, fullscreen and in a window.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

Scene 3, OpenGL 3.3 core, 45 s.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.

Scene 4, OpenGL 3.3 core, 45 s.

Medians of the fresh sessions of alternating from 2026-11-10 23:09 to 11-11 07:30, Hypoland master r957 (after 0.3.0) against Hyprland 0.57.1-2.