NullDC4Wii - Dreamcast Emulator for Wii

a fork from https://github.com/skmp/nullDCe

TODO (Maybe you can help !)

Simple

• Test current state with every game and report compatibility (see "compatibility" below)
• Create presets for games
• Test 2/3/4 player mode on wiimote & gamecube also, please report
• Help me finding regression (NOT bugs or glitch, only regression for now please)
• Comment / Guides / Documentation (WiiBrew Wiki)
• Test and report Fishing Rod/USB Keyboard/Lightgun/Maracas support

Developer (Easy)

• Controller correct layout, for pro pad and for gamecube pad
• User custom Preset
• Fishing Rod/USB Keyboard/Lightgun/Maracas support (probably unsupported now)
• Put external config file for controllers (controls.cfg)
• fix Volgarr regresion
• DualShock 3 issue: left stick has the Y axis inverted, up is down, and down is up.
• Custom layout : Chuchu rocket/Quake3 (bith Gamecube & wii)
• Custom layout : a toggle to make the right analog stick with supported controllers act as buttons X (Left) - Y (Up) - A (Down) - B (Right). This is useful for twin stick shooters that use the diamond button layout for gameplay (like Xeno Crisis). (like chuchu probably ? so it's done I think)
• Add a winCE preset to put an additional menu message (after option screen) to prevent that game is WinCE and isn't supported (wince=yes in game_presets.cfg). Message be like : "This is a WinCE game, it's not supported yet by NullDC4Wii (and probably never will). Press A to launch Anyway, B to return to file selection."
• Add zlib-compressed CHDs Support (cdzl). Make a message that CHD with cdlz is not supported

Developer (Normal)

• 4/3 support (implemented, need fix on some games like Shenmue)
• Support for CHD/ELF game file
• Return to file list (instead return to homebrew menu)
• Synchronise Dreamcast clock to Wii's clock
• Get RGB565_opaque_alpha AUTO mode
• Get Offset color AUTO mode

Developer (Hard)

• Fix Chuchu Rocket 16Bit PCM bug
• Fix non WinCE games not launching (Rez, San Francisco 2049...)
• Improve gxRend.cpp = main file about specific rendering for Wii
• Splitting gxRend.cpp in multiple files ? (beware this is more tricky than it look)
• Table convertion between SH4 Opcodes of SH4 and the WiiPPC ?
• Use LLVM to port code for PowerPC ? (skmp says its not a good idea in this case)
• Dynarec improvement (very performant right now, but if we can find boost...)
• WinCE Games support #37

Installation

Put BIOS file and game file

Mandatory BIOS files in SD:/data/

• dc_boot.bin
• dc_flash.bin
• fsca-table.bin (included)

Optional BIOS files in SD:/data/

• dc_flash_wb.bin (this is the dc_flash but already saved)
• syscalls.bin (needed for elf/bin)
• IP.bin (needed for elf/bin)

dc_nvmem.bin
vmu_default.bin

Game file in SD:/discs/ or USB:/dreamcast/

Test with castlevania Resurrection and Sega Tetris to begin with

Put your folders with GDI in this directory. CDI also works

Might work for ISO / BIN / CUE / NRG / MDS

BIN/CUE/ELF, but you probably need IP.bin/syscalls.bin (take IP.TMPL from bootdreams and rename it IP.Bin)

Configuration

General configuration

Check nullDC.cfg at root

Controls

Dreamcast	Wiimote	Wiimote (ChuChu Rocket!)	Gamecube	Gamecube (ChuChu Rocket!)
A	A	down & A	A	A
B	B	right & B	B	X
Y	1	up	Y	Y
X	2	left	X	B
START	Home	Home	START	START
D-PAD	D-PAD	- no implementation -	D-PAD	D-PAD
STICK	Nunchuck Stick	Nunchuck Stick	STICK	STICK
L	- (and Nunchuck Z)	- (and Nunchuck Z)	L	L
R	+	+	R	R
To Exit	- and +	- and +	R + L + Z	R + L + Z

Exit :
Press - and + (wiimote) or Press L + R + Z (or L + R + Start)

Special layouts (CONTROLS menu)

The last menu before launch has a SPECIAL LAYOUT row. It is not saved - it goes back to OFF on every boot, like every other menu setting.

Value	What it does
OFF	Normal per-device mapping (the table above)
CHUCHU ROCKET	The ChuChu Rocket columns of the table above
DDR SELECT	Dance Dance Revolution Club Mix / 2nd Mix: holding the Gamecube Z button reads as "analog stick pushed fully down", which is how those games take a Select input (dancer / arrow skin / sequence type, and the Left, Right and Shuffle option codes)
USER CFG	The whole mapping comes from user_controls.cfg instead - see below

A game preset can set this row for you with layout=off|chuchu|ddr_select|user_cfg in game_presets.cfg. [default] sets layout=off, which is what resets the row between game selections - do not delete that line, or a layout set by one game stays active for every game you launch afterwards in the same session.

user_controls.cfg (USER CFG layout)

Fully remappable controls, from a file you edit by hand. Put user_controls.cfg next to boot.dol or in your games folder - same places game_presets.cfg is looked for - then set SPECIAL LAYOUT to USER CFG. The file itself documents the format and lists every target and every physical source you can bind.

The CONTROLS menu tells you whether it was actually found: with USER CFG selected the row reads (user_controls.cfg LOADED) or (user_controls.cfg NOT FOUND). If the file is missing, USER CFG quietly falls back to the normal mapping, so it can never lock you out.

Note: the user_controls.cfg that ships with the emulator deliberately reproduces the default mapping line for line. Selecting USER CFG without editing it first will look like nothing happened - that is expected. Edit the file, then reboot the emulator: it is read once at startup.

Sixaxis / DualShock 3 (USB) - hold Gamecube B at boot

DS3 support is off by default and has to be switched on at boot: hold B on a Gamecube controller (any port) while the emulator starts. The CONTROLS menu shows SIXAXIS/DS3 (USB): [OFF] when it was not switched on.

It has to be opt-in because raw USB access to the pad needs IOS58, and switching the console to it restarts the whole I/O system - Bluetooth included. Every connected Wiimote drops its link and has to reconnect right as the emulator is starting up, which is why Wiimotes would sometimes refuse to connect in the menu or in game. Leave the button alone and none of that happens.

It has to be a Gamecube button, not a Wiimote one: the Gamecube ports are read directly by the console's main CPU, so they work before any of this is decided, while reading a Wiimote would need the very Bluetooth stack in question - and no Wiimote has connected yet that early in boot anyway.

A console with no Gamecube ports (Wii Family Edition, Wii Mini, Wii U vWii) can therefore not enable DS3 support.

VMU (Memory card)

It seems to be supported, but 1rst you'll need to format the VMU in the bios

Files appears at root of /data/ :

• vmu_save_A1.bin
• vmu_save_A2.bin

Compatibility

https://wiibrew.org/wiki/NullDC4Wii/Compatibility

Presets

Presets are grouped in the in-emulator menu across 7 pages. The order below follows the exact same order shown on screen (Page 1 to Page 7). "Default" below is what actually ships out of the box, from game_presets.cfg's [default] section (applied on every launch before any per-game section) - see game_presets.cfg further down for how to override any of this per game.

Page 1 : General

RATIO

Mode	Settings	Rendering
ORIGINAL	4/3 pillarbox	Original Dreamcast ratio
FULLSCREEN (default)	Stretched to fill screen	Fullscreen
AUTO	Picks by console aspect ratio setting (4:3 console → full width, 16:9 console → pillarbox)	Depends on console setting

SPEED LIMITER

Mode	Settings	Rendering
OFF (uncapped, default)	Emulator can run above 100% speed	Uncapped
ON (cap 100%)	Stops speed exceeding 100%	Capped

SHOW FPS

Mode	Settings	Rendering
OFF (default)	No overlay	Nothing displayed
ON	Displays gameplay FPS and speed overlay	Overlay shown

🖼️ Graphics Preset

OLD Behavior (Before alpha0.64) :

Mode	Settings	Best platform
LOW	GX_NEAR · lod_bias 0.0f · GX_DISABLE	Wii
NORMAL (default)	GX_LINEAR · lod_bias 0.0f · GX_DISABLE	Wii
HIGH	GX_LINEAR · lod_bias -0.5f · GX_ENABLE · Anisotropic x2	Wii U
EXTRA	GX_LINEAR · lod_bias -0.75f · GX_ENABLE · Anisotropic x4	Wii U

The visual difference is limited for NORMAL/HIGH/EXTRA

NEW behavior (from alpha 0.64) :

Mode	Settings
LOW	GX_NEAR
NORMAL (default)	GX_LINEAR

• Use LOW for 240p games/modes
• Use NORMAL for other games

Important note : LOW can cause Z-Fighting (example in jet set radio, see #115)

The old HIGH/EXTRA levels were nothing but NORMAL plus a fixed lod_bias/bias-clamp/anisotropic bundle - those are now the three separate presets below (GX / LOD BIAS / ANISO), so any combination is possible instead of only the 4 fixed tiers.

GX (LOD extras)

Mode	Settings	Rendering
OFF (legacy)	GX_DISABLE on GX_InitTexObjLOD's biasclamp + edgelod	Standard
ON (default)	GX_ENABLE on biasclamp + edgelod	biasclamp stops LOD_BIAS pushing a minified texel past the point its footprint no longer covers the pixel; edgelod computes LOD from adjacent instead of diagonal texels

ANISO forces this on by itself, since libogc requires edgelod whenever anisotropy > 1.

LOD BIAS

Mode	Settings	Rendering
-1.0 / -0.75 / -0.5	Sharpen: samples a larger mip level than the footprint asks for	Sharper, can shimmer
0.0 (default)	Hardware default	True no-op
+0.5	Blur	Softer

ANISO

Mode	Settings	Rendering
0x (off, default)	No anisotropic filtering	Standard
2x / 4x	Anisotropic filtering (Hollywood tops out at 4x - no 8x on this GPU)	Sharper on steep-angle surfaces, costs fill rate

Requires MIPMAP set to FAST or TRILINEAR (Page 6) - does nothing while mipmaps are off, since anisotropy is only iterated with a GX_LIN_MIP_LIN min filter.

TEXTURE CACHE

Mode	Settings	Rendering
VERY_FAST	skmp original algorythm (magic numbers). Buggy in most games	Max FPS
VERY_FAST+	very_fast's address-derived slots, but oversized textures route into a properly-sized arena instead of overrunning the next slot (3 targeted fixes - stride surfaces, stride sizing, mip sentinel). Try this first when very_fast is the only fast option but shows corruption	Max FPS, safer than VERY_FAST
FAST	Best performance/accuracy in most case	Almost Max FPS
NORMAL (default)	Display mostly correctly	Good FPS
QUALITY (SLOW)	Best accuracy. Display correctly	Mid FPS

Can have huge FPS impact, try to have the lowest parameter.

VQ CMPR

Mode	Settings	Rendering
OFF (default)	VQ textures decode to 16bpp	Standard
ON	VQ textures decode straight to GX CMPR/DXT1 (4 bits/texel)	Pairs with TEXTURE CACHE VERY_FAST/VERY_FAST+: a VQ texture now fits its address-derived slot instead of overrunning its neighbour - the classic VERY_FAST VQ corruption. Second lossy pass on top of VQ, so smooth gradients can band; 565 VQ only

FRAMESKIPPING

Mode	Settings	Rendering
0 (default)	No frame skipped	Every frame drawn
1	Skip 1 frame	Faster, less smooth
2	Skip 2 frames	Faster still, less smooth
AUTO	Skips frames while behind real time, min 1 render in 4	Adaptive, holds emulated time at 100%
AUTO_MAX	Same as AUTO, but up to 9 skips in a row	Speed over framerate

The FPS counter shows the RENDER rate, so a low FPS with SPEED at 100% is working as intended.

2D FRAMEBUFFER

Mode	Settings	Rendering
NO (default)	Disable	Standard rendering
YES	Enable 2D framebuffer	Try for 2D games

Still on testing. debug_log_framebuffer2d (Page 6 debug logs) tells you whether a given game would even take this path before enabling it.

ADVANCED_ALPHA

Mode	Settings	Rendering
NO	basic alpha threathment	Not accurate
YES (default)	additionnal alpha threatment	Near perfect

Mostly for debug. Should always be on

> BLEND_MODE

Mode	Settings	Rendering
OFF (LEGACY)	Disable	Not accurate
ON (CORRECT) (default)	Activate BLEND_MODE	Accurate, correct for Resident Evil 3

If flickerings, try turning off

Note : ADVANCED ALPHA needs to be on for BLEND_MODE

>> FPS_BOOST

Mode	Settings	Rendering
OFF (CORRECT) (default)	Disable	Accurate
ON (FASTER)	+2 FPS in BLEND MODE, to the cost of wrong Alpha/Transparency	Not Accurate

note : ADVANCED ALPHA and BLEND_MODE needs to be on for FPS_BOOST

PUNCH THROUGH

Mode	Settings	Rendering
OFF (legacy, default)	PT polys drawn last in TR blend state	Faster, less accurate
ON (correct)	OP → PT → TR order + PT_ALPHA_REF alpha test	Correct PT list alpha test

Needed in lot of games

TRANS_SORT

Mode	Settings	Rendering
OFF (default)	Disable	not Accurate (faster)
ON	can display stuff	Accurate

Can resolve flickering in some games. Needed in lot of games. Superseded by AUTOSORT (Page 3) for intersecting/interleaved geometry that a painter sort alone can't fix.

RENDER TO TEX

Mode	Settings	Rendering
OFF (faster, default)	RTT frames dropped (legacy)	Faster, mirrors/TV screens missing
ON (correct)	EFB copied back into VRAM	Correct mirrors/TV screens
OVERLAY (carry)	Pass not resolved as a texture - its geometry is carried into the next display frame and drawn last, flat on top (parked on the near plane, GX_ALWAYS, no Z-write)	For passes the game composites itself as an overlay rather than as a texture, e.g. Silent Scope's sniper crosshair
KEEP	ON, plus the render does NOT consume the TA list, matching real hardware (the ISP/TSP just walk the tile arrays, which stay valid until TA_LIST_INIT)	For a game that renders one accumulated list twice with different write addresses/clip windows, e.g. Silent Scope's scope disc + 24x-magnified world. Pair with SPLIT SCREEN so the magnified half stays inside its own tile clip

Needed in some games. With OFF, a dropped pass's geometry still leaks into the next frame (drawn first, misclassified as opaque) - that's why some overlays half-show even with this off.

SPLIT SCREEN

Mode	Settings	Rendering
OFF (faster, default)	Every render pass presented fullscreen (legacy)	Faster
ON / TILE CLIP	Both viewports in ONE render pass, each carrying a PVR User Tile Clip rect	Correct 2P viewports, e.g. Daytona USA
MULTI-PASS	One RENDER_START per viewport, into its own band of the EFB; ONE assembled frame is presented	Fixes heavy 2P flicker (player1/player2/player1 alternating) in Le Mans 24 Hours, Demolition Racer, Magical Racing Tour
BOTH	Per-poly tile clips inside multi-pass renders	Combination of the two above

Needed for 2 players splitscreen or any 2 camera angle games.

Page 2 : Graphics

FMV FORMAT

Mode	Settings	Rendering
CMPR (DXT1)	Compressed format	Use if some movie displays white
RGBA8	Uncompressed, full quality	Slower
RGB565	Uncompressed, no alpha	Faster
TEV (default)	No CPU colour math at all: the YUV422 source uploads as raw I8 luma + IA8 chroma planes and the BT.601 matrix runs in the GameCube/Wii TEV combiner instead	Fastest, chroma gets hardware bilinear the CPU paths never had; 8-bit TEV coefficients drift ~1-2 LSB from the CPU matrix

YUV STRIDE

Mode	Settings	Rendering
OFF	Always decode at the texture's declared (power-of-two) width	Slices a non-padded FMV frame into repeated strips
AUTO	Use the real per-row pitch only for a texture the YUV converter actually wrote	Fixed Bomberman's intro, Test Drive 6, Dino Crisis
ALWAYS	Take the pitch from TA_YUV_TEX_CTRL for EVERY YUV422 texture	Old behaviour; a game that never programs that register gets it decoded as 16x16 (Soul Calibur's character select went mostly black)
TEXCTL (default)	AUTO, plus TCW.StrideSel (TEXT_CONTROL[4:0] x 32 texels) as a fallback for surfaces AUTO's converter-address tracking never saw (it only remembers 4 base addresses)	Also fixes the planar RGB path. If a movie is still messy under AUTO, try this before ALWAYS

YUV TWIDDLE FIX

Mode	Settings	Rendering
OFF (legacy)	Twiddled YUV422 textures decode with luma/chroma swapped	Green<->magenta duotone with 1-column striping on affected art
ON (default)	Correct luma/chroma extraction for twiddled YUV422	Fixes static YUV artwork such as Virtua Fighter 3tb's "FIRST MATCH" loading screen

Static YUV artwork only, NOT movie playback - real (planar) FMV is unaffected either way.

Vertex Color

Mode	Settings	Rendering
OFF (grey scale)	Grey Scale	Grey scale (a tiny bit faster)
ON (default)	Intensity color	Accurate

Color some pixel (Used in Jet Set Radio Future and Crazy Taxi 1/2)

SPRITE COLOR

Mode	Settings	Rendering
OFF (legacy)	Every sprite's Base Colour field discarded, drawn as hardcoded white	A black sprite with ModulateAlpha shading draws white instead
ON (default)	Honours a Sprite's own packed Base Colour (all 4 corners)	Correct - found via Fighting Vipers 2's SEGA screen, whose full-screen black backdrop is a sprite that drew as a white plate

Pair with VTX ALPHA for sprite-based fades.

VTX ALPHA

Mode	Settings	Rendering
OFF (legacy, default)	Vertex alpha forced to 0xFF on every ARGB1555 polygon regardless of TSP.UseAlpha - a deliberate hack for ARGB1555 cutout fonts (Test Drive 6)	Keeps cutout fonts correct
ON	Honours TSP.UseAlpha as the hardware does	Needed by anything that FADES via vertex alpha on an ARGB1555 surface, e.g. Fighting Vipers 2's SEGA screen full-screen fade

DECAL_ALPHA

Mode	Settings	Rendering
OFF (faster, default)	no decal alpha	Not accurate
ON (correct)	Decal alpha implemented	Accurate, fixes Crazy Taxi's cars

See more : #68

SEAM FIX

Mode	Settings	Rendering
OFF (a bit faster, default)	Disable	Thin black seam lines between 2D tiles/sprites remain
ON	Half-texel UV inset	Fixes black lines between 2D tiles

Use this or LOW graphics to fix seam lines. See #18

Warning : ON causes a bug with Vertex Displacement (water mostly) : #119

FOG

Mode	Settings	Rendering
OFF (legacy, default)	TSP.FogCtrl decoded but never applied	Nothing ever fogged
ON	Per-polygon PVR2 fog honouring TSP.FogCtrl: LUT (FOG_TABLE/FOG_DENSITY), per-vertex (offset-colour alpha), or LUT mode 2, evaluated per vertex and blended by one extra TEV stage	Distance haze instead of geometry popping in at the draw distance - racers/outdoor games mostly

Costs 4 bytes/vertex + one TEV stage on fogged polygons only, and recolours every fogged polygon in the scene - stays per-game.

BG POLYGON

Mode	Settings	Rendering
OFF (faster, default)	v0 color used for EFB clear only, no background quad drawn	Faster
ON (correct)	Barycentric-extrapolated background quad drawn	Correct bg gradient/texture, e.g. Who Wants to Be a Millionaire

Caused an FPS regression in other games when left unconditionally on - enable per-game only.

RGB565 ALPHA

Mode	Settings	Rendering
OFF (fmt0 only, default)	Only fmt0 (ARGB1555) forced opaque	Correct for POD 2
ON (fmt0+fmt1)	Force opaque for fmt0(ARGB1555)+fmt1(RGB565)	Turn off for POD 2

May disapear in a future

JOJO FIX

Mode	Settings	Rendering
OFF	Disable	Pre-fix behavior
ON (default)	TLUT-reupload-skip + CACHE_FAST PalSelect-masking	For JoJo's Bizarre Adventure

Has to be used with CI4_FAST/CI8_FAST to reduce massive FPS drop in battle. May use the same technique in other games.

OFFSET COLOR

Mode	Settings	Rendering
OFF (legacy)	Offset/specular color dropped	Standard
ON (default)	PIX = base*tex + offset via 2nd TEV stage	Correct specular highlights

Costs 4 bytes/vertex of FIFO + a second TEV stage on offset polys. May cause white surface on some games (ie = berserk, Tokyo highway challenge...). May fix black surface on some games (ie = sega worldwide soccer)

4BPP MODE / 8BPP MODE

Mode (4BPP/8BPP)	Settings	Rendering
I4_STUB/I8_STUB	Dummy algorythm	Some element doesn't display at all, for max FPS
OPTIMIZED	Served as test, in the end CI4/CI(FAST) is better	Very good FPS
CI4 (FAST)/CI8 (FAST) (default)	Best performance/quality	Very good FPS
CI4 (NORMAL)/CI8 (NORMAL)	Advanced algorythm for CI4/CI8	Mid FPS
RGB565 (ACCURATE)	Most advanced algorythm	Can have massive FPS dropdown (1 FPS) on some games

Page 3 : Depth & Width

DEPTH_CLIP

Mode	Settings	Rendering
OFF (legacy)	XF Z-clipping on, no near margin	2D/menus can be invisible on real Wii
NEAR MARGIN (default)	Pads vtx_min_Z 0.1% so the nearest 2D layer can't land exactly on the near clip plane	Recommended for Wii - CONFIRMED, fixes ChuChu Rocket and Crazy Taxi menu/intro with no downside seen
NO CLIP (Dolphin)	Matches Dolphin: out-of-range depth clamps instead of the poly vanishing	Regressed Crazy Taxi when combined with FIXED DEPTH=TIGHT - avoid that combo

It's basically like FIXED_DEPTH, leave it to NEAR MARGIN

FIXED_DEPTH

Mode	Settings	Rendering
OFF / DYNAMIC (default)	Legacy per-vertex min/max W tracking, fits Z range to the scene each frame	Good
WIDE	Skips that tracking, fixed planes W=[0.0001..100000] - safe everywhere, coarser Z	can help display some stuff - mostly for debug, more Z-fighting risk
TIGHT	Fixed planes W=[0.1..25000] - much finer Z, but geometry outside that range clips	can help display some stuff

FIXED_DEPTH can help flickering and Z-Fighting

HUD_PASS

Mode	Settings	Rendering
OFF (default)	-	Not active
OVERLAY (no Z-write)	Help hud to display when FIXED_DEPTH is on TIGHT	accurate, but may be overdrawn by geometry drawn after it
PROTECT (Z-write at near plane)	Help hud to display when FIXED_DEPTH is on TIGHT	Perfect - use if OVERLAY leaves polys in front of the HUD

Mostly needed if Fixed Depth is set to tight. A no-op on its own (dynamic/wide ranges never clip the HUD). Fixes the "Z-fighting gone but HUD vanished" case in Rayman 2 / Cannon Spike.

LEGACY DEPTH

Mode	Settings	Rendering
OFF (default)	Current depth pipeline	Standard
ON	Reproduces the depth pipeline exactly as it stood at commit 1bb8c27: fixed planes NEAR=0.001/FAR=10000*1.001, vert_base 1/W clamp at 0.001 (not 0.0001), no per-vertex min/max tracking or margin/HUD fixups	For a game that only rendered correctly at that old commit - Buggy Heat's logo/VMU screen/gameplay. Overrides FIXED DEPTH; the 0.001 clamp is the one thing FIXED DEPTH cannot express

Shown first on this page in-game, right after LAUNCH.

SUBPASS ZCLEAR

Mode	Settings	Rendering
OFF (legacy, default)	Single shared depth pass - a later pass inherits whatever the main scene left in the Z buffer	Standard
ON	Re-parks the whole Z buffer at a known W via a full-canvas GX_ALWAYS quad (colour untouched) right before a later geometry group	Gives e.g. a HUD PASS=PROTECT overlay a clean depth baseline instead of the main scene's leftovers

PPZ_WRITE : PER POLYGON Z WRITE

Mode	Settings	Rendering
NO	No Per Polygon Z Write	More compatible
YES (default)	Per Polygon Z Write	More accurate

Try putting NO if you experience troubles, with HUD for example.

ISP_DEPTH_FUNC

Mode	Settings	Rendering
OFF (legacy, default)	Disable	Standard
ON (opaque/PT)	Per-poly depth test on opaque/PT lists, honouring each polygon's actual isp.DepthMode register instead of the fixed GEQUAL painter compare (translucent list stays GEQUAL, matching real PVR autosort)	Experimental
ON (all lists)	Same, applied to all lists	Experimental

Different from LAYER SORT below: this reads the hardware's own per-polygon compare mode instead of a heuristic sort.

ISP_CULL

Mode	Settings	Rendering
OFF (legacy, default)	Disable (never cull)	Standard
ON	Per-poly backface cull from isp.CullMode	Experimental
ON (swap winding)	Per-poly backface cull, two cullable windings swapped	Use if plain ON makes geometry vanish / look inside-out

AUTOSORT

Mode	Settings	Rendering
OFF (legacy, default)	Disable	Standard
1-4 layers (slow)	Real per-pixel PVR autosort via GX depth peeling, value = max translucent depth layers composited per pixel	Stronger than TRANS_SORT's per-strip painter sort, for intersecting/interleaved translucent geometry. Very GPU-heavy (~2 extra TR walks + 2 EFB Z copies per layer) - use 2 or 3 only where needed, per-game only. Best paired with PUNCH THROUGH=ON

Overrides TRANS_SORT; LAYER SORT overrides this.

LAYER SORT

Mode	Settings	Rendering
OFF (legacy, default)	Disable	Standard sort
ON (TR tier sort)	Layer-tiered translucent sort	For 2D scenes drawn at ONE depth

Helps determine what should be front and back trough looking at texture format/properties. For games that submit their whole 2D scene at a single depth and rely on the Dreamcast's per-pixel autosort: background plates draw first, then stage art, then stage sprites, then everything else. Game-agnostic — used by Hokuto no Ken and Street Fighter III. Was part of the HOKUTO HACK before alpha0.66.

LIST ORDER

Mode	Settings	Rendering
OFF (legacy, default)	Flat strip buffer walked in submission order	Assumes the game submits its opaque list first
ON	If the opaque list opened AFTER the translucent one, its range is drawn first, as real hardware would	Fixes "gameplay visible for a moment, then a big background image covers everything" - Puyo Puyo 4/DA!, whose two full-screen background plates go out in the OPAQUE list after the gameplay sprites

Bit-identical to OFF for any game that submits OP first (condition never fires).

X SCALER

Mode	Settings	Rendering
OFF (legacy)	PVR SCALER_CTL.hscale support disabled	Only the LEFT HALF of the image shows in affected games
ON (default)	PVR SCALER_CTL.hscale support	For Omicron / Wacky Races (render 1280 wide, scaler halves 2:1)

For Nomad Soul and Wacky Racer. Maybe other games

Y SCALER

Mode	Settings	Rendering
OFF (legacy, default)	SCALER_CTL.vscalefactor ignored	Only a SLICE of the image shows in affected games
ON	Scales the projected canvas height by that factor (6.10 fixed point, 0x400=1.0 - above 1.0 the CORE renders taller and the scaler shrinks on write / vertical SSAA, below 1.0 the opposite)	Whole scene shows - needed for Silent Scope's sniper scope to be 100% accurate

The vertical counterpart of X SCALER - same register family, other axis.

H SCALER

Mode	Settings	Rendering
OFF (legacy, default)	VO_CONTROL.pixel_double ignored	In low-res video modes the framebuffer holds a HALF-width (320) image doubled by the video DAC - scene is drawn in a 320-wide screen space, filling only the LEFT HALF of a 640 canvas
ON	Halves the projected canvas to match	Whole scene shows - mandatory for several games, and needed for 2-player splitscreen in some

Register-driven counterpart of CANVAS WIDTH below, which stays the manual override for games that render narrow WITHOUT setting the bit. An explicit CANVAS WIDTH wins over this.

CANVAS WIDTH

Mode	Settings	Rendering
OFF (640, legacy, default)	Legacy 640 canvas	Standard
Custom value	Forces canvas width in 240p modes	e.g. SF3 Double Impact = 384 (the CPS3 arcade width)

See compatiblity wiki for more info

POLY OFFSET

Mode	Settings	Rendering
OFF (default)	No bias	Legacy
Tier 1-3	Native polygon offset / Z bias via a Z-texture in ADD mode (GX_SetZTexture), applied to the Punch-Through list - increasing bias strength	GX equivalent of the co-planar decal/shadow/road-marking sort that real PVR tile order used to give for free (real hardware has no glPolygonOffset-style register)

Page 4 : Audio

AUDIO BUFFERS

Mode	Settings	Rendering
DEFAULT (saved)	Leaves the value at its cfg/UI setting	Depends on saved config
0 (never block)	Never blocks/drops on overrun	Fastest, most likely to drop
1 / 2	Blocks until below N queued buffers	More paced
3 (most paced)	Most conservative pacing	Most paced, safest

Put audio buffers = 1 generally leads to good audio. To the cost of FPS unfortunatly.

CDDA MUSIC

Mode	Settings	Rendering
OFF	CD audio tracks silent	No CD music
ON (default)	GD-ROM Red Book audio fed to the AICA mixer	CDDA music plays in games

MUTE 16BIT PCM

Mode	Settings	Rendering
OFF (legacy, default)	All AICA sample formats audible	Standard
ON (silence 16B)	16-bit PCM channels silenced at KEY_ON	Fixes ChuChu Rocket's echoey 16-bit SFX (also mutes any other 16-bit music/voices, so game-specific)

AICA FAST

Mode	Settings	Rendering
OFF (legacy)	All 64 AICA channels visited every sample, even idle ones	Standard
ON (default)	Only playing channels are visited, empty filter-envelope call skipped, no DSP send while the DSP is off, master volume cached, silent CD-audio skipped	Same output samples, cheaper mixer - Wii-CONFIRMED: aica% 10.0->7.7, speed +2.6%, no audible change

ARM7 SPEED

Mode	Settings	Rendering
10MHZ (default)	Sound CPU at normal speed	Standard audio
5MHZ (faster)	Underclocked sound CPU	Faster, check audio!
2.5MHZ (risky)	Heavily underclocked sound CPU	Risky, check audio!

5 mhz generally works and bring FPS boost

ARM7 JIT

Mode	Settings	Rendering
OFF (interpreter)	Cached ARM7 interpreter	Standard audio
ON (basic)	ARM code translated to PPC, same results/timing as the interpreter	Faster, Wii-confirmed
ON (linked, default)	Same, plus blocks branch straight to each other instead of going back through the dispatcher after every branch	Fastest, Wii-confirmed

A conformance test suite runs once at ARM init (see [ARM7JIT] in /ndclog.txt); a failure falls back to the interpreter automatically. Compare arm:% on the stats line and listen before keeping a change.

ARM7 SPEED and ARM7 JIT both govern the sound co-processor, so they moved here from Page 5 (Core) to sit next to the other audio presets.

Page 5 : Core

🧮 Calculation Accuracy Preset

Mode	Description
FAST (default)	Maximum FPS (higher frame rate), less loading times
BALANCED	Good balance between speed and accuracy
ACCURATE	Closest behavior to original hardware

If you experience Freeze in some heavy games like Shenmue, put FAST or BALANCED. FAST may be the default setting in future versions

If you experience various bugs (example that may happens : weird AI controled NPC, weird timing) put ACCURATE

ASYNC_RENDER

Mode	Settings	Rendering
OFF (legacy)	CPU blocks in GX_DrawDone until the GPU finishes each frame	0 frame latency
ON (faster, default)	Frame queued, presented one vblank later; SH4 emulates while GPU draws	Faster, to the cost of 1 frame input-lag

ASYNC_RENDER is generally faster. Can resolve flickering. Works better on real hardware than dolphin

RENDER DELAY

Mode	Settings	Rendering
OFF (faster, default)	Legacy: instant list-complete IRQs	Faster
ON (hw-like)	Hardware-like staggered ISP/TSP/Video timings	ON for MvC2 and CvSNK

TMEM CACHE

Mode	Settings	Rendering
OFF (legacy)	Full GPU texture cache invalidate every frame	Standard
ON (faster?, default)	Invalidate only on texture re-decode	Keeps GPU texture cache warm

Haven't seen any effect but keeping on for now

SH4 CLOCK

Mode	Settings	Rendering
200MHZ (full, default)	Nominal Dreamcast speed	Full speed
Underclock (150-200MHz, step 5)	Lower value = fewer emulated cycles per real second	Lower = faster host, slower game

Underclocking is supposed to raise FPS. Didn't see any difference

Page 6 : Experimental Stuff & Debug

MIPMAPS

Mode	Settings	Rendering
OFF (fastest, default)	No mip chain	Fastest, more shimmer far away
FAST	Generated GX mip chain + nearest-mip bilinear	Less shimmer far away
TRILINEAR (slow)	Best quality	Best quality, halves texture fill rate (e.g. -40% in Test Drive 6)

Required for ANISO (Page 1) to have any effect.

TEX WRAP GUARD

Mode	Settings	Rendering
OFF (legacy, default)	Persistent texture arena wraps without draining the GPU first	If the GPU is still sampling old textures when the 14 MB cache arena wraps, their bytes get overwritten mid-sample - arena-wrap race corruption
ON	Mid-frame, if the arena fills, calls GX_DrawDone() to drain the GPU queue before wrapping; a second exhaustion in the same frame falls back to the skimp slot	Wii-CONFIRMED mechanism (watch [TEXC] drains=/wraps=); did NOT fix the JSR intro streaking it was tried against, so it's not a general streak fix

TEX CLAMP FIX

Mode	Settings	Rendering
OFF (legacy, default)	A cached texture keeps the FIRST polygon's ClampU/V+FlipU/V wrap mode forever, since those are per-POLYGON TSP bits baked into GXTexObj only on decode	A tiled floor inheriting GX_CLAMP smears its edge texels to infinity - "streaked"/"infinite X or Y" symptom
ON	Every texture bind re-applies the binding polygon's own wrap mode via GX_InitTexObjWrapMode before GX_LoadTexObj	Wii-CONFIRMED: fixes streaked textures in Crazy Taxi (VERY_FAST+ trees), Deadly Skies (FAST/VERY_FAST+ floors), Jet Set Radio (VERY_FAST+ intro). Does NOT fix Shadow Man's sky (QUALITY cache re-decodes every frame, never susceptible)

DMA FIX

Mode	Settings	Rendering
OFF (legacy)	Disable	Standard
ON (default)	ch2/PVR/Sort/AICA-G2 DMA correctness fixes ported from the verified-working NullDC PSP port	Fixes related to loading CDI/GDI file

SCHED (ORDER)

Mode	Settings	Rendering
OFF (cascade, default)	Legacy Medium/Slow/VerySlow timeslice cascade	Standard
ON (deadline)	Unified cycle-deadline scheduler: GD-ROM read-done, ch2/PVR/AICA-DMA completion, render-done and TA list-end fire through one deadline queue in true hardware order	Experimental, related to loading CDI/GDI file

EXIT FIX

Mode	Settings	Rendering
OFF (legacy)	Bare exit(0)	Can leave a black screen with the Wii still on when returning to the Homebrew Channel
SAFE SHUTDOWN (default)	Ordered teardown before exit	Fixes the black-screen hang
SAFE + WATCHDOG	Safe shutdown + loader-stub watchdog	For a hang the plain safe shutdown doesn't catch
SAFE + WATCHDOG + HOST POLL	Same, plus polls the exit combo host-side once a frame	For a game that stopped polling Maple, so the exit combo itself can still fire

TRANS ZWRITE (debug only)

Mode	Settings	Rendering
ON (default)	Legacy: the TR list inherits the frame-wide GEQUAL compare WITH depth write, so the first translucent strip submitted near depth stamps the Z buffer and depth-rejects farther translucent strips behind it	One near full-screen quad can hide the rest of the list
OFF	Keeps the depth TEST (opaque geometry still occludes translucent behind it) but drops the write, so translucent strips composite in submission order	DEBUG ONLY: hides the Dreamcast BIOS boot logo, which needs the depth ordering this removes

Survives purely as an investigation tool for scenes where translucent strips wrongly occlude each other.

HOKUTO HACK

Mode	Settings	Rendering
OFF (legacy, default)	Disable	Standard sort
ON (HnK addresses)	Hardcoded RAM address hack	Specific fix for Hokuto no Ken

Refines LAYER SORT using Hokuto no Ken's own VRAM texture addresses, for the debris tiles that texture format/properties alone cannot tell apart from the fighters. Needs LAYER SORT on — it does nothing by itself — and only works for stage 1/2 at the moment. Leave it off in every other game.

PUYO HACK

Mode	Settings	Rendering
OFF (default)	Disable	Standard
ON	Folds Puyo Puyo 4's own two hardcoded backdrop VRAM addresses into the LAYER BACK TEX tier-4 sort	Fixes the gameplay playfields (puyos going dim behind the grid) and the intro/main screen background (fading UI/logo pieces buried under it)

Puyo Puyo 4 ONLY. A menu-row twin of the cfg-only layer_back_tex key (see game_presets.cfg below) so the fix works without touching the SD card at all - e.g. from inside Dolphin.

DINO CRISIS INVENTORY FIX

Mode	Settings	Rendering
OFF (default)	Disable	Standard
ON (redecode)	Per-game hardcoded texture redecode hack (icon slot)	Dino Crisis inventory fix

Hardcoded to one address - meaningless for any other game.

SH4 CORE

Mode	Settings	Rendering
INTERPRETER	Interpreted SH4 core	Slow, for debugging only
DYNAREC (default)	JIT recompiler	Fast

Debug logs

All default OFF, all write to /ndclog.txt on the card, costing a card write every frame while on - turn back off once a log has answered its question.

Row	What it logs
Dbg FB2D Log	Whether a game ever takes the bit-24 render path 2D FRAMEBUFFER (Page 1) would act on - answers "is it worth trying on this game?" before flipping it
Debug Message	Renderer trace ([PATH], [FB], [RTT]...)
Debug Loop	Per-loop CPU/GD-ROM/IO trace. VERY slow
Debug GDROM	GD-ROM / CDDA SPI command trace
Debug Skip Tex	Hides every polygon using one texture (VRAM address, hex or decimal, from a [SCN] census addr= field). Diagnostic only - it REMOVES geometry, never fixes anything. Answers whether an element is missing because something draws OVER it, or because it's drawn wrong itself - this is how the Hokuto no Ken layering bug was pinned down

Page 7 : JIT/DYNAREC

Perf and diagnostic presets for the SH4 dynarec and its interpreter-fallback path. Several read at COMPILE time, not per-frame - set them before launching, not mid-game.

JIT SBP

Mode	Settings	Rendering
OFF	No stale/self-modified block guard	Fastest, riskiest
KNOWN (default)	Guards known self-modifying regions (DOA2LE, Shenmue 1/2, ...) plus the boot-entry cache flush	Balanced
ALL RAM (slow)	Guards all RAM	Safest, slowest, diagnosis only

FASTMEM

Mode	Settings	Rendering
OFF (legacy)	Standard PPC-MMU memory access	Slower
ON (faster, default)	Branchless JIT MMU-mapped memory access via segment regs + hashed page table; MMIO/SQ/BIOS accesses DSI-fault once and back-patch to slow-path trampolines	Faster

Crash observed in Re-Volt when launching a race. Only game that does that for now.

JIT BCACHE

Mode	Settings	Rendering
OFF (legacy)	Dynamic jumps (jmp/rts/bsrf) chase cache[] -> DynarecBlock across two cache lines + a counter write	Standard
ON (flat, default)	One flat, 1-cacheline {addr, code} dispatch entry per dynamic jump	Faster dispatch. L1/L2 cache related - can help heavy scenes like Shenmue's intro

JIT DYN IC

Mode	Settings	Rendering
OFF	No inline cache on dynamic branch exits	Standard
ON (JSR/JMP)	Per-site inline cache on JSR/JMP dynamic exits	Experimental
ON (+RTS, default)	Same, plus RTS	Wii-measured +0.98% - the whole win is in RTS sites; JSR prediction actually runs backwards, so there isn't much more to gain here

FPU PIN

Mode	Settings	Rendering
OFF (legacy)	fr0-15 not pinned	Standard
ON (default)	Pins SH4 fr0-15 to real PPC FPU registers f14..f29 for the whole session	Speeds up geometry-heavy games (fadd/fmul/fmac/fipr/ftrv/cvt_* stop round-tripping through memory)

JIT ALIGN

Mode	Settings	Rendering
OFF (legacy)	No block alignment	Standard
ON (32B lines, default)	Pads every SH4-dynarec block entry to a 32-byte Broadway L1 cache line	Cache-hygiene only, no logic change, marginal effect

JIT IFB FLUSH

Mode	Settings	Rendering
OFF (full spill)	Every interpreter-fallback opcode (28 of them, incl. div1, addc, subc, mac.l, tas.b...) gets a FULL register-file spill - 15 stores + 15 loads, +32 more with FPU PIN on	Correct, slow
ON (selective, default)	Narrows the spill per opcode to only the GPR/float registers that opcode actually touches, via a closed allow-list	Same result, cheaper. div1 is the costly one - SH4 has no divide instruction, so a 32-bit software division is ~32 div1 calls, each paying the spill

An opcode not on the allow-list keeps the full spill either way - slow, never wrong. Best candidates: games doing lots of integer division. Pair with JIT IFB PROBE to see if a game issues enough to matter.

JIT IFB PROBE (diagnostic)

Mode	Settings	Rendering
OFF (default)	No counting	Standard
ON (logs [IFB])	Counts interpreter fallbacks per opcode, writes an [IFB] breakdown to /ndclog.txt once a second: total/s, how much JIT IFB FLUSH could narrow, spill memory-op rate with/without it	Costs 4 instructions per fallback site - turn off for timing runs

Read at compile time - set before launching.

JIT NEW OPS

Mode	Settings	Rendering
OFF (legacy)	9 opcodes (stc SR, stc.l SR, ldc.l SR, ldc SR, lds FPSCR, lds.l FPSCR, rotcl, rotcr, tas.b) always fell back to the interpreter even though dynarec code for them already existed - it just wasn't reachable from the opcode table	Full interpreter call-out + register spill every time
ON (9 ops jitted, default)	Wires all 9 into the dispatch table	stc SR alone measured up to 96k/s in ChuChu Rocket scenes; all 9 now skip JIT IFB FLUSH's allow-list entirely

Changes SH4 codegen - read at compile time. Run JIT IFB PROBE first to see if a game issues enough of these.

JIT HOTBLOCKS (diagnostic)

Mode	Settings	Rendering
OFF (default)	No counting	Standard
ON (logs [HOT])	Counts executions per compiled SH4 block, writes the hottest ones to /ndclog.txt once a second as [HOT] lines with codegen density (PPC bytes/SH4-opcode), plus a full SH4+PPC dump of the top block twice per session	Costs 4 instructions per block entry - turn off for timing runs. Covers everything JIT IFB PROBE doesn't (i.e. non-fallback codegen)

Read at compile time - set before launching.

JIT T-FORWARD

Mode	Settings	Rendering
OFF (legacy)	A compare followed by bt/bf (the commonest branch shape on SH4) emits stw T then lwz T back to back - a load-hit-store stall on Broadway to recover a value that never left the register	Standard
ON (no T reload, default)	Forwards the T bit straight from the register it was just computed in to the branch that consumes it - only when the compare is the immediately preceding op; the store stays since a later block may read T	Found via JIT HOTBLOCKS on the #1 block of a ChuChu scene

Changes SH4 codegen - read at compile time.

JIT FMOV DIRECT

Mode	Settings	Rendering
OFF (GPR bounce)	fmov reads/writes bounce through a GPR and memory	Standard
ON (LFS/STFS, default)	fmov reads/writes straight to/from the pinned FPRs on the fastmem path	+10% Wii-CONFIRMED (Castlevania 103%→113%). Needs FPU PIN + FASTMEM both on - this "Phase B" of FPU PIN had been disabled in every real run until fixed

JIT CARRY OPS

Mode	Settings	Rendering
OFF (legacy)	addc/subc/negc/addv/subv/div1/cmp-str/xtrct/swap.b/clrmac always call the interpreter	Standard
ON (10 ops jitted, default)	Real SHIL ops mapping T onto PPC XER[CA]/XER[OV]	With JIT IFB FLUSH on, a fallback is ~35-45 cycles; expect tenths of a percent, not a breakthrough (~0.3% of frame in Castlevania). Also a correctness fix: the interpreter's div1 had the wrong Q^M^carry step

JIT MAC OPS

Mode	Settings	Rendering
OFF (legacy)	mac.l/mac.w call the interpreter, including their memory reads	Standard
ON (fastmem reads, default)	Their @Rm+/@Rn+ reads become ordinary readm ops through FASTMEM instead of the interpreter's ReadMem; mullw/mulhw + addc/adde inlined	Also makes SR.S=1 saturation mode work - it used to be fatal for mac.l and silently skipped for mac.w

Expect little effect for most games - the SH4 has a real FPU, so MAC is rarer here than on the SH2 this technique came from.

JIT FSCHG FAST

Mode	Settings	Rendering
OFF (legacy)	fschg (toggles FPSCR.SZ only) still calls the full UpdateFPSCR() - a C call + 16 lfs + 16 stfs just to keep old_fpscr in step	Standard
ON (1 stw, default)	One stw instead	DC's vertex T&L brackets its loop with fschg, firing ~2.3x per transformed vertex. Wii-CONFIRMED +10% in ChuChu mouse mania at matched load; null in Castlevania (doesn't run this hot)

JIT CR0 BRANCH

Mode	Settings	Rendering
OFF (legacy)	A conditional block exit rebuilds T via mfcr/rlwinm/stw/cmpi	Standard
ON (default)	Branches directly on the PPC CR0 bit the compare already set (sr_T is still written, since a later block may read it)	Wii-measured +2.8% on BIOS boot phase; null in Castlevania gameplay. Only fires when the flag producer is the block's LAST op

JIT CCALL CENSUS (diagnostic)

Mode	Settings	Rendering
OFF (default)	No counting	Standard
ON	Counts every C function the dynarec still calls out to, prints a [CC] block once a second (per-second AND per-frame rates: scheduler tick, interrupt, bm_GetCode, do_sqw/TA, UpdateSR, UpdateFPSCR, fsqrt, fsrra, ReadMem/WriteMem misses, mac saturation)	Far cheaper than JIT HOTBLOCKS since counters live inside the callees, but still re-measure speed with it off before trusting a comparison

The census that found writemem at 1.1M calls/s in Crazy Taxi (57% of all JIT call-outs), which led to BLOCKCOPY below.

BLOCKCOPY

Mode	Settings	Rendering
OFF (legacy)	Store-queue flushes and PVR/ch2 DMA transfers run the memory dispatcher once per 32-bit word	Standard
ON (default)	Bulk copy instead	Wii-CONFIRMED +2%. Crazy Taxi measured 1,104,636 system-RAM writes/s, 95.6% from plain C code rather than JIT code. MMIO, mirror wraps and region crossings still take the per-word path

JIT FSQRT

Mode	Settings	Rendering
OFF (legacy)	fsqrt calls newlib sqrtf - a 25-iteration shift-and-subtract bit loop bracketed by two load-hit-store stack round-trips (~300 cycles)	Standard
ON (default)	Inline frsqrte + 3 Newton-Raphson steps + an exact-residual step, verified bit-identical to sqrtf over 2.07M simulated cases	Wii-CONFIRMED +3.5-4.3% (Crazy Taxi, which measured 167,906 fsqrt/s). Zero, negative, Inf and NaN still take the old libm call - frsqrte alone is only a 5-bit estimate and once distorted the BIOS swirl

Game Specific Presets

LAYER SORT (Hokuto No Ken, Street Fighter III) helps determine what should be front and back trough looking at texture format/properties. HOKUTO HACK adds to it a RAM address check for a few Hokuto no Ken textures that those properties cannot separate — Hokuto no Ken only, and it needs LAYER SORT on.

JOJO FIX (Jojo's Bizarre adventure) has to be used with CI4_FAST to reduce FPS drop in battle. May use the same technique in other games

Vertex color is a special method on Dreamcast to color stuff. Notable example in Crazy Taxi : The arrow and dollar sign are Vertex Colored. You will notice the cars too : They will add to the current texture so the same car with the same texture will appear red, blue, green, etc...

Same for Jet Set Radio, will add color to the logo.

LEGACY DEPTH (Buggy Heat) reproduces one specific old commit's depth pipeline verbatim, for a game that only ever rendered correctly at that point in history.

PUYO HACK (Puyo Puyo 4 / DA!) and its cfg-only twin layer_back_tex/layer_front_tex (see game_presets.cfg below) fix 2D games that submit a backdrop or overlay plate at the same depth as the sprites it should stay behind/in front of - painter order alone can't tell them apart.

wince=yes in game_presets.cfg is not a rendering preset - it just shows a warning after the options menu ("this is a WinCE game, not supported yet - A: launch anyway, B: back to file list") instead of silently failing to boot, for the handful of Dreamcast games built on Sega's Windows CE devkit instead of plain Katana.

See Compatiblity guide for hints depending of the games

game_presets.cfg

game_presets is a file that reads the file name and directly apply matching presets.

game_presets.cfg needs to be in sd:/discs/

With recent version, multiple file names are suppported :

[crazytaxi2][crazy taxi 2] ; more specific first
depth_clip=1
tex_cache=normal
vertex_color=off

[crazytaxi][crazy taxi]
depth_clip=1
tex_cache=normal
graphics=normal

So in that configuration, your file name should be crazytaxi.gdi or crazytaxi.gdi

I'm not enterely sure spaces are supported right now

Maybe it needs to be more specific, so in this example, just add [crazytaxi1] like this :

[crazytaxi][crazytaxi1][crazy taxi]

Then rename your gdi crazytaxi1.gdi or crazytaxi1.cdi

If everything is correct, in the option menu, you should see something like this :

Creating the optimal Preset :

Best thing to do :

1/ Launch with default preset
2/ play a little bit the game (default stage, just input A A A), 2/3 minutes is enough
3/ If some things doesn't display good, try to change settings. Already we know that works :

• fine (seam) lines in 2D sprites : put "LOW" as graphics
• Z fighting : put FIXED DEPTH to "tight"
• 2 player viewport (or 2 cameras in the same scene) = SPLITSCREEN

4/ Flickering, fix can be :

• BLEND_MODE (set to off)
• TRANS_MODE (set to off)
• ASYNC_RENDER (set to on)
• DEPTH-CLIP (set to tight)

5/ Other graphical things not displaying (or black screen) : try flipping all the other setting, particulary :

• 4BPP/8BPP to CI4_FAST or CI8_FAST
• BG POLYGON to on
• PUNCH Trough to ON
• TRANS_SORT to ON
• RENDER TO TEX to ON
• 2D Famebuffer = ON

6/ off Canvas problem (screen too small or too big)

• try xratio = on (or off)
• CANVAS_WIDTH (try different values)

7/ Restart the game, try CACHE_VERY_FAST and CACHE_FAST. Find the lowest compatible option for the game. They may crash heavy-scene thus, like Shenmue intro

8/ Try if FPS_BOOST is possible and don't break too much the game

9/ If you want to dig more, try the game in early version of NullDC4Wii :

• alpha0.25
• alpha0.28
• alpha0.40

Report to me if you seen any regressions

Some examples of common problems :

Fine (seam) lines in 2D example :

CACHE related problem examples :

Off Canvas :

Z Fighting example :

Fastmem crash example :

For Developpers :

Compilation Process (Windows)

0/ Download/clone source code

1/ Install devkitpro/devkitPPC

https://wiibrew.org/wiki/DevkitPPC

Just tick PPC (not ARM, x86, etc).

See this issue : #13

2/ Launch MSys2 terminal

Devkitpro has it's own UNIX terminal, by default it's located here :
C:\devkitPro\msys2\usr\bin\mintty.exe

3/ Install additional development packages :

pacman -Syu # updates MSYS2 and package database
pacman -S wii-dev

4/ PATH & System variable configuration (Windows)

PATH

In windows variable environnement add C:\devkitPro\devkitPPC\bin to Uservariable PATH

UPDATE MARCH 2026 : This folder seems to be needed also for elf2dol : C:\devkitPro\tools\bin

System variables

Modify these system variable

DEVKITPPC : C:\devkitPro\devkitPPC
DEVKITPRO : C:\devkitPro\

Strongly advise you to completly reboot Windows after that (not just relaunching CMD)

5/ launch wii/vs_make.bat in a standard CMD windows terminal

Correct errors if they are some errors

Use dollz3

dollz3 is a compress tool for *.dol files, and it is in the original "vs_make.bat" file, but it seems not to work

https://wiibrew.org/wiki/Dollz

Compilation Process (Linux & Mac)

It should work for Linux & Mac with similar process

Leaving this link for now :

https://wiibrew.org/wiki/DevkitPPC

Dolphin (for debug/testing)

Activate :

• SD Card
• Display FPS
• For log, in config.ini add "DebugModeEnabled = True" under [Interface]
• (optional) VSync eventually
• (optional) Advanced > Debug > Texture Format Overlay

Frequently Asked Question (FAQ)

My SD Card isn't recognized

Make sure you have a good brand (Samsung, Kingston). Optimal card would be between 2Gb and 32Gb (more should work, we support SDXC), formated in FAT32 with 32kb cluster (16kb or 64kb could work). Try another SD card if that's not working

Games doesn't work

First, try on SD card with a know working game (Castlevania, Chuchu Rocket or Sega Tetris are best candidate).

My USB key/HDD isn't recognized

For USB, try a FAT32 USB key first (32kb cluster) before switching to any other device. It works on my 500HDD FAT32 for information (with a cluster of 4kb)

Does VMU work ?

It's reported working. Step 1 : go to bios. Step 2 : format VMU. Step 3 : exit properly the emulator (+ and -)

Why this emulator ?

Because why not ? At first it was a POC (Proof Of Concept) but digging more and more, it seems somes games are achievable to 100% speed (in PAL50 mode). So yeah !

What games does work ?

Please check the compatiblity Wiki : https://wiibrew.org/wiki/NullDC4Wii/Compatibility

Does it work on Wii U ?

We have a Wii U fowarder that also use the speed boot of the Wii U (overclock). A native port for Wii U could be done if someone wants to do port it.

Does it support Atomiswave ?

Not yet, this could ask between 0.03 and 0.5Mb of RAM. Doable yes, maybe not really worth the effort. But you can use Atomiswave hacks for dreamcast, lots of them do work.

Will it support NAOMI ?

Definitly not, the recent Dynarec/JIT (= fps boost) and ARM7 cache ask for more RAM, we have no more place to add 16Mb for NAOMI Ram

Controls are messy / this doesn't work / My Propad isn't recognize etc...

Open an issue, everything should be ok now

Will THAT AMAZING ADVENTURE GAME be considered as "Supported" one day ?

Probably not, this emulator prioritize multiplayer fun games to play friends with. Playing an adventure game like Shenmue or an RPG could be a very frustrating experience that I strongly do not recommand. Use Gamecube/Wii version (if it does exist) or a PC for emulation instead.

What can I do to help ?

Try different games, parameters and report in the compatibility wiki : https://wiibrew.org/wiki/NullDC4Wii/Compatibility

Is 100% speed of the emulator achievable on Wii ?

The only way to know this answer is to try our best ! Actually, some 2D games run 100% speed, and some 3D games also ! But we are mostly around 60 or 80% of speed...

We are still trying to improve performance, altough we already done a lot of amazing stuff :

• Improved Dynarec
• Fastmem
• ARM7 CACHE

We also need to work more on TEV (Texture Environnement - a specialized hardware unit inside the graphics chip), it has been tested but only on dolphin. Maybe on real hardware this would have better result.

Another good strategy would be to really have a per-game specific emulator. That would take a long time, so for now we have presets & auto-presets implementation trough game_presets.cfg

Also, it's suggested we build a custom Dolphin-Emulator with some tools to help fine-tuning everything and gain more speed. More info here : #116

Is 100% speed of the emulator achievable on Wii U?

Probably ! Wii U has the additional CPU power we need. Test the overlocked fowarder !

Can it read Original games (GD-Rom) ?

The Wii can't read those disc format. Maybe if you change the optical drive and with some code implementation, but that's not worth it.

Can it read CDI/Utopia disc ?

Maybe it's possible, but we don't have time to focus on this. CDI Files on SD Card/USB are supported since alpha 0.28 anyway.

Will WinCE games be implemented ?

That an additionnal ressources in CPU and we are limited. That may would make sense for a WiiU Port.

We are currently testing a branch with some implementation but it doesn't seem to work.

Will Retroachievement be implemented ?

Probably not, we are super tight on RAM, and it also cost 1/2% of cpu cycle

Will Netplay be implemented ?

Probably not, again, we are very short on RAM/CPU. And on Wii we don't have the 100% speed everytime.

How is AI involved in the project ?

Since I (BenoitAdam) digged the NullDC code, AI has been heavily used to make improvement to the emulator. The very first state of the emulator (alpha 0.02) is 99% hand written code by SKMP and NullDC contributors at the time. Only some few changes have been made to be able to recompile it and make it run. Various AI are used : MistalAI/ChatGPT/Codex/Claude and Deepseek. Claude is very convenient because of artifact and Claude Code. Gemini helped on some improvements, and Deepseek too.

I hate AI !

It's ok, you have the right, but without AI this project wouldn't have been resurected. AI for code is really a big help, definitly not the same thing with AI generated images and videos. For information AI for image ask ~10x more power, AI for video ask ~100x more power. A Wii is also using 20x less power than a PS4/PS5.

Reduce CO2 emission & grow trees is the plan for the planet. Also prevent stupid people throwing their cigarett butt & firmly condemn pyromaniacs.

Do you have a discord ?

Yes : https://discord.gg/sJst6jmQyH

Do you accept donation ?

Yes ! Initially I was against donation but spending this much time & effort (and sometimes money) to this project led me to ask for donation. Not trying to getting rich, just staying afloat.

Buy Me a Coffee / Ko-Fi

This is very popular amongst small dev project, I don't know what's the difference or if some people would prefer one over the other one, anyway here are the links :

• https://buymeacoffee.com/nulldc4wii
• https://ko-fi.com/nulldc4wii

Patreon

The classic "Patreon" subscribing is here :

https://www.patreon.com/cw/NullDC4Wii

Paypal

For those used to Paypal, the link is here : https://www.paypal.com/ncp/payment/2NE7AYS77K6AJ

Ressources

Dreamcast Emulators

NullDC https://code.google.com/archive/p/nulldc/source/default/source
NullDC (github) https://github.com/skmp/nullDCe
NullDC for PSP : https://github.com/PSP-Archive/nulldce-psp
NullDC for Xbox360 https://github.com/gligli/nulldc-360
Reicast : https://github.com/skmp/reicast-emulator
Flycast : https://github.com/flyinghead/flycast
Deecy : https://github.com/Senryoku/Deecy Redream : http://redream.io/

Devkitpro

Main website : https://devkitpro.org
​GitHub : https://github.com/devkitPro
Installer (releases) : https://github.com/devkitPro/installer/releases
Wii ​Examples : https://github.com/devkitPro/wii-examples

libOGC

GitHub (Wii/GameCube system librairy) : https://github.com/devkitPro/libogc

Emulators​

Dolphin Official Website : https://dolphin-emu.org
GitHub (for timings Gekko/CPU) : https://github.com/dolphin-emu/dolphin

Wiibrew Wiki

Emulation Page : https://wiibrew.org/wiki/Emulation
Homebrew tutorials : https://wiibrew.org/wiki/Main_Page

Media Coverage

France :
https://www.programmez.com/actualites/programmez-numero-gaming-et-developpement-de-jeux-est-disponible-39720

Italy :
https://www.biteyourconsole.net/2026/05/13/nulldc4wii-alpha-v0-22-porta-il-dreamcast-su-wii-con-supporto-usb-multiplayer-e-miglioramenti-prestazionali/

Credits

• skmp (original NullDC creator)
• NullDC contributors
• Joseph Jordan - libiso
• Xale00 (also know as Benoit Adam) - 2026 recompilation
• gligli (Xbox 360 port)
• Xeihro (Xiro28 PSP Port)
• AI because this project would have never existed otherwise lol
• Probably Reicast/Flycast Team also
• Welcome to the IA-age guys and good luck everyone.

All together, let's Cast the Dream.

Special thanks

• skmp because he's the god
• Senryoku, develloper of Deecy emulator
• OriginalDave, developer of Jocasta
• MetalliC, developer of Demul
• Dolphin Team, would have been a pain to test on real hardware everytime otherwise
• devkitPPC/Libogc
• Everyone helping on emuvdev/discord
• Jilou04 for constantly testing on Wii U and reporting
• All testers and all futur testers
• People actually helping me on the wiibrew wiki

Special no thanks

• To all people not believing in this project
• People constantly critisizing the fact AI is used in this project (this project wouldn't have existed otherwise...).