OpenRCT2

Played sessions

Nobody has played it on Walnut yet: no session, no medal.

Scripted benchmarks

Not rated: tests of driver and game changes. A script plays the same scene on the PS3 again and again (the desktop stopped, vsync off) to compare two builds of the game or of the graphics driver. The numbers say which build is faster; they never feed the medal.

How the script plays it: OpenRCT2 0.5.6 (our big-endian build, gamedb/games/openrct2/packages/openrct2/: openrct2 0.5.6+ds-1+ps3.N, plus openrct2-data and openrct2-music, both arch all; Debian's is little-endian only) with the user's RollerCoaster Tycoon 2 data (roller-coaster-tycoon2-data in /usr/share/games/roller-coaster-tycoon2, or the old rct2-data in /usr/share/games/rct2): smoke test, not benchmarked. The software renderer (the OpenGL engine needs GL 3.3): the game draws an 8-bit frame on the CPU and SDL's renderer (opengl on nouveau) scales and presents it. Base: borderless fullscreen 1920x1080 with window_scale 2, so the game renders 960x540 and SDL doubles it (the setting to ship: the PPE paints a quarter of the pixels). Variant full: window_scale 1 (renders 1920x1080). The park loads straight from the command line, no input needed: Electric Fields (small map, rides and guests from the start); variant big: Six Flags Magic Mountain (the largest RCT2 park; 2782 guests after 10000 ticks in our oracle run). Variant title: no park, the title sequence. Variants nosound (sound, ride music and title music off) and lean (nosound plus multithreading off and MALLOC_ARENA_MAX=1: the RAM knobs measured on the PC), sdlsw (SDL's software renderer and no GL framebuffer: no Mesa in the process at all). The simulation runs at its fixed 40 ticks per second; FPS counter on (top middle of the screen; the frame rate is capped at 40 too, uncap_fps off). The park's date at the bottom right should advance. Settings: a private user directory, $HOME/.config/OpenRCT2-ps3bench (--user-data-path; the player's own ~/.config/OpenRCT2 stays untouched) with config.ini written by setup() (BACKUP restores it; the game rewrites it at exit). infer_display_dpi=false (else the first start replaces window_scale with the display's DPI ratio), last_run_version set (no changelog window), autosave never (no disk writes during the run), game_path=the data folder (the game also finds either itself). First start only: the game indexes its objects (2533 from objects 1.8.0 plus RCT2's ObjData), scenarios and tracks into the user directory: on the PC 0.1 s, on the PPE expect seconds to a minute. Risks: RAM. Measured on the PC (amd64 build of the same source, Xvfb 1920x1080, SDL software renderer without GL in the process; RSS median over 120 s after a 60 s warm-up, build/tmp/openrct2 ram.sh): Six Flags Magic Mountain (2782 guests): 309 MB at 1920x1080, 300 MB at 960x540 x2, 260 MB lean (anon 228); Electric Fields: 264 / 253 / 232 MB. Of that, 70 MB is the game state: the entity table is sized for 65535 entities (512 bytes each, 57.7 MB for the registry; RCT2 parks hold at most 10000), zero-filled at start. With SDL's opengl renderer the GL driver comes on top (llvmpipe on the PC: +140 MB; nv30 is less, and the 960x540 texture is VRAM). Expect swap: watch MemAvailable, swap-ins and stalls. Speed: the PPE paints the viewport every frame (dirty rectangles only when nothing moves), unknown; the sim alone under qemu runs ~10x slower than the PC.

Conditions: what a variant adds to the recipe (its environment switches) and what ran alongside (the stall recorder, a profiler). Length: "warm-up + 120 s" is a steady-state run, which leaves out a 60 s warm-up and measures 120 s or more; "whole run" counted everything from the start, the way runs before 2026-10-17 did. Rows marked invalid or not comparable say why.