Latency, Two Guns End to End, Troubleshooting, and Where Every Game Runs

6.1 Where the time goes on Windows

The whole chain from hand to picture, and why a fixed delay is forgiven while a variable one is felt, are worked out in Aiming Maths and Calibration, Volume 7, section 7.5, and the cable’s share in Fully Wired, Volume 5. The gun’s share is about 2.7 to 8.8 ms, 5.6 ms on average with averaging on, including the up-to-1 ms wait for the PC’s USB poll. Windows itself adds almost nothing to the input: Raw Input delivers the report as it arrives; what Windows and the emulator decide is how long the picture takes.

The model in the chart, for a game running at 60 frames per second (one frame = 16.7 ms):

Table 1 — The model in the chart, for a game running at 60 frames per second (one frame = 16.7 ms)

Stage	Composed window, double-buffered, 60 Hz	Flipped (full screen or borderless, Independent Flip), 60 Hz	Variable refresh, panel up to 144 Hz
Gun’s share, including USB poll	5.6 ms	5.6 ms	5.6 ms
Game samples input once per frame (average)	8.3 ms	8.3 ms	8.3 ms
Game logic and rendering (one frame)	16.7 ms	16.7 ms	16.7 ms
Waiting in the presentation queue	16.7 ms (one frame)	8.3 ms (average wait for the next refresh)	about 0 (the panel refreshes when the frame is ready)
Scan-out to mid-screen	8.3 ms	8.3 ms	3.5 ms (half of 6.9 ms)
Total before the display’s own processing	about 56 ms	about 47 ms	about 34 ms

The numbers are illustrative: they are frame periods and the gun’s design figures, not a measurement, and many games take two or three frames of internal lag, not one. But the differences between the columns are real and come from the frame arithmetic. Two points stand out. Getting the presentation right is worth one to two frames, more than the whole gun. And variable refresh helps an emulator twice: the frame no longer waits for a fixed refresh tick, and a 144 Hz panel scans each 60 Hz frame out in 6.9 ms rather than 16.7 ms, so the middle of the screen, where the targets are, updates about 5 ms sooner. Most gun games ran on 60 Hz arcade monitors, so a VRR panel runs them at exactly their own rate with none of the judder a fixed 144 Hz refresh would add.

6.2 Presentation: composed, flipped, or variable refresh

Microsoft’s guidance for developers explains the middle column. With the flip presentation model, “windowed mode [is] effectively equivalent or better when compared to the classic ‘fullscreen exclusive’ mode”: when a window covers the screen, Windows can skip desktop composition and send the program’s frames “directly to the screen, independently, with the same efficiency as fullscreen exclusive”, a mode called Independent Flip, in which “you can get down to 1 frame of latency”. An older blt-model program in a window is composed by the desktop window manager and pays the extra queueing in the first column.

Three practical consequences for a gun on Windows:

• Borderless full screen is fine in any program that presents with the flip model. Whether a given emulator does was not established program by program for this dive; the Direct3D 11/12 and Vulkan back ends that MAME (bgfx), RetroArch, Flycast, DuckStation and PCSX2 offer are the likely candidates. Which mode is actually in use is shown by Microsoft’s PresentMon tool, which the same page recommends.
• Variable refresh (G-Sync, FreeSync, HDMI 2.1 VRR) is enabled in the graphics driver and the display, then per program: RetroArch’s “Sync to Exact Content Framerate (G-Sync, FreeSync)”, and MAME’s lowlatency, which “is particularly effective with VRR (Variable Refresh Rate) displays”.
• Television game mode. A television’s picture processing can exceed everything in the table above. Its game or PC mode is the single most important setting, and it must be measured rather than assumed (6.4).

Exclusive full screen remains available in most programs and does no harm, but on current Windows it is no longer the only route to the lowest latency, and borderless modes make Alt-Tabbing to a hook program’s window painless.

6.3 The settings, program by program

Table 2 — 6.3 The settings, program by program

Program	Settings for low latency with a gun
MAME 0.289	lowlatency 1 (“draw a new frame before throttling to reduce input latency”; may jitter in some 3D systems, hence off by default); waitvsync 1 with a fixed-refresh display, syncrefresh only for games at the display’s rate; video d3d or bgfx
RetroArch v1.22.2	VRR sync on; Max Swapchain Images 2 (1 where allowed); Frame Delay as high as the PC allows; Run-Ahead where the core supports it; Polling Behavior Late (Volume 3, section 3.7)
Flycast, DuckStation, PCSX2	Each has a “Vsync” and a frame-pacing or low-latency option in its graphics settings; use VRR where available. Their exact option names were not compared for this dive
TeknoParrot games	Run the game’s own renderer; use the profile’s windowed or full-screen setting and VRR in the driver
Everything	Crosshair off once the gun is calibrated (Volume 2, section 2.9): it makes the lag visible without reducing it

6.4 Measuring it on the bench

The end-to-end number is easy to get with the lab’s Analog Discovery Pro ADP3450: one channel on the trigger switch (or the solenoid gate, which the firmware fires on the trigger), another on a fast photodiode taped to the screen where a game flashes or a target reacts. A shooting gallery game that flashes the screen on every shot (the Model 2 games do; DemulShooter’s Lua scripts can remove that flash, Volume 4) gives a clean edge. Fifty shots per configuration give a histogram whose median is the latency and whose width is the variable part.

To separate the display from the PC, the same measurement with the photodiode replaced by the GPU’s own output (a frame counter on a capture card, or the display’s input-lag figure from its own test pattern) leaves the PC’s share. Aiming Maths and Calibration, Volume 7, section 7.6 isolates the gun’s share from the PC entirely.

6.5 Two guns, end to end

Every program in this dive tells the guns apart its own way. The one decision that serves all of them is to give each gun a distinct identity, and then to do each program’s step once:

Table 3 — Every program in this dive tells the guns apart its own way. The one decision that serves all of them is to give each gun a distinct identity, and then to do each program's step once

Where	Player 1 gun	Player 2 gun	Pinned against replugging?	Volume
OpenFIRE App, USB identity	Preset P1: PID 0x0001, “FIRECon P1”, Start 1 / Select 5	Preset P2: PID 0x0002, “FIRECon P2”, Start 2 / Select 6	Yes (in the gun)	1
Windows Device Manager, COM port	COM1 (for MAMEHooker)	COM2	Per USB port	1
MAME	mapdevice VID_F143&PID_0001 to GUNCODE_1	…PID_0002 to GUNCODE_2	Yes	2
RetroArch	Port 1 Mouse Index = FIRECon P1	Port 2 Mouse Index = FIRECon P2	No: check after any change of port	3
Supermodel	MOUSEn mappings with -input-system=rawinput	the other n	By configuration order	4
Flycast, DuckStation	Port A / Port 1 bound to the P1 gun	Port B / Port 2 to the P2 gun	Not checked	4
PCSX2	Mouse (pointer 0)	Gamepad mode, Relative axes	—	4
TeknoParrot	P1 LightGun bound to FIRECon P1	P2 LightGun to FIRECon P2	By device path	5
DemulShooter	P1 Config page	P2 Config page	By device	5
QMamehook	First gun detected	Second	Detection order	5

Two habits remove most two-gun trouble: plug both guns in before starting any program, and keep each gun on the same USB port, ideally the PC’s own ports rather than a hub. MAME’s own documentation says that devices not connected at start-up “may not be numbered as expected”, and every program that stores a number (RetroArch, Supermodel) behaves the same way.

6.6 Troubleshooting

Table 4 — 6.6 Troubleshooting

Symptom	Most likely cause	Fix	Where
Aim is fine on one monitor, wildly wrong on a multi-monitor PC	Absolute coordinates map to the primary monitor	Make the gun’s screen primary	1.4
Correct at the centre, drifts outward toward the sides	The game is pillarboxed but the gun covers the whole screen	Stretch the game, or M3x1 from the hook’s start command	1.7, 2.3
Desktop cursor right, game off by a constant amount	The arcade game’s own calibration	Recalibrate in its test menu; in Demul, quit cleanly so NVRAM saves	4.1
Two guns move one cursor, or only one gun works	The program is reading the system cursor	Use its Raw Input path; in RetroArch, raw driver and a non-SDL2 video driver	1.3, 3.2
Two guns work but swap players after a reboot	Enumeration order changed	MAME mapdevice; elsewhere, re-check the port mapping; keep guns on fixed ports	2.6, 6.5
RetroArch or TeknoParrot confused by two guns	Both guns have the same product ID	App presets P1 and P2	1.5
The gun is invisible in a TeknoParrot profile	Input API set to XInput	Set Input API to RawInput	5.2
No recoil in a MAME game with a hook running	Wrong output setting for the hook, or the ini key is not the driver’s output name	output windows for MAMEHooker, network for QMamehook; let the hook write the ini’s names first	2.10
MAMEHooker shows a DLL error	Not run as administrator	Run as administrator (OpenFIRE wiki), or use QMamehook / OutputHooker	5.7
DemulShooter does not hook the game	Not administrator; renamed executable; started in the wrong order	Run as administrator; -pname=; start the hook, then DemulShooter, then the game	5.3, 5.4
MAMEHooker cannot find the gun	Gun is COM5 or higher	Renumber to COM1–COM4 in Device Manager	1.6
Off-screen reload does nothing	The gun pins at the screen edge, so the game never sees off-screen	MAME plugin; RetroArch Gun Reload; Supermodel InputAutoTrigger; off-screen trigger mapped to right click for DemulShooter games	2.8, 3.6, 4.3, 5.4
The solenoid stays engaged during a machine-gun burst, then cuts out	Hold-type output with F0x1; the firmware’s 2,000 ms limit	Map the “on” state to a pulse command instead	2.10, 5.6
Recoil fires twice per shot	The gun’s own trigger recoil and the hook’s both firing	The hook must send S at start (the gun’s LED turns white); and an ini line without a bar fires on both the “on” and “off” changes	OSF Vol 6 §6.7, 5.6
Aim jumps erratically in parts of the screen	Reflections or a stray infrared source	The App’s IR camera view shows a fifth dot	OSF Vol 5 §5.4

6.7 Where it runs: the game-by-game table

What each project’s own documentation and source show for each game: the Windows route, whether a second gun has a documented path, and where game-driven recoil comes from. “Documented” means the route is in the program’s own documentation or code, not that it has been played with this gun; nothing has been, yet. Recoil sources are DemulShooter’s Outputs table (custom Recoil column), MAME’s driver source, or “gun’s own” where the game provides nothing and the gun fires on its trigger.

Table 5 — 6.7 Where it runs: the game-by-game table

Game	Original platform	Windows route	Second gun	Game-driven recoil	Notes
Terminator 2: Judgment Day	Williams Y-unit	MAME	Yes (mapdevice)	MAME: Left_Gun_Recoil, Right_Gun_Recoil	Hold-type recoil: map to pulses
Alien3: The Gun	Sega System 32	MAME	Yes	MAME: Player1_Gun_Recoil	ini\source\segas32.ini
Revolution X	Midway X-unit	MAME	Yes, and a third	MAME: three Player*_Gun_Recoil
Beast Busters	SNK	MAME	Yes, and a third	MAME: three Player*_Gun_Recoil
Point Blank 2	Namco System 12 / 11	MAME	Yes	MAME: Player*_Gun_Recoil (System 12) or recoil0/1 (System 11 sets)
Ghoul Panic	Namco System 12	MAME	Yes	MAME: Player*_Gun_Recoil
Under Fire	Taito	MAME	Yes	MAME: P1_gun_recoil, P2_gun_recoil
Lethal Enforcers	Konami	MAME	Yes	None found in the driver; gun’s own	Off-screen reload plugin, named in its documentation
The House of the Dead	Sega Model 2	Model 2 emulator + DemulShooter	Yes	DemulShooter: Yes (genuine and custom)	Calibrate in test menu
Virtua Cop, Virtua Cop 2	Sega Model 2	Model 2 emulator + DemulShooter	Yes	DemulShooter: Yes	Named in MAME’s reload-plugin documentation too
Behind Enemy Lines	Sega Model 2	Model 2 emulator + DemulShooter	Yes	DemulShooter: custom Yes, genuine No	Middle click = grenade
Gunblade NY	Sega Model 2	Model 2 emulator + DemulShooter	Yes	DemulShooter: Yes
Rail Chase 2	Sega Model 2	Model 2 emulator + DemulShooter	Yes	DemulShooter: custom No	”aiming is…weird”
The Lost World	Sega Model 3	Supermodel, Raw Input	Yes	None (Supermodel has no gun output); gun’s own	InputAutoTrigger for one-click reload
The House of the Dead 2	Sega Naomi	Flycast (Raw Input) + DemulShooterX64, or Demul + DemulShooter	Yes	DemulShooter: Yes
Confidential Mission	Sega Naomi	Demul + DemulShooter (Flycast listed in DemulShooter’s usage)	Yes	DemulShooter (Demul): Yes
Death Crimson OX, The Maze of the Kings, Lupin the Third: The Shooting	Sega Naomi	Demul + DemulShooter; Flycast for Maze and Lupin	Yes	DemulShooter: Yes
Ranger Mission, Sports Shooting USA, Extreme Hunting	Sammy Atomiswave	Demul 0.7a + DemulShooter	Yes	DemulShooter: No	Edge offset unless 4:3 window or DX11 plugin
The House of the Dead III, Virtua Cop 3, Ghost Squad	Sega Chihiro	Cxbx-Reloaded + DemulShooter	Yes	DemulShooter: Yes
The House of the Dead 4, 4 Special	Sega Lindbergh	TeknoParrot + DemulShooter	Yes	DemulShooter: Yes	TeknoParrot OffScreenReloadHack option
The House of the Dead: EX	Sega Lindbergh	TeknoParrot + DemulShooter	Yes	DemulShooter: custom No, genuine Yes
Let’s Go Jungle, Special	Sega Lindbergh	TeknoParrot + DemulShooter	Yes	DemulShooter: Yes	TeknoParrot: Intel GPUs “NO”
Too Spicy, Rambo	Sega Lindbergh	TeknoParrot + DemulShooter	Yes	DemulShooter: Yes
Sega Golden Gun, Let’s Go Island, Transformers: Human Alliance	Sega RingWide	TeknoParrot + DemulShooter	Yes	DemulShooter: Yes
Operation G.H.O.S.T., Target Bravo	Sega RingWide	TeknoParrot + DemulShooter	Yes	DemulShooter: custom Yes (no Damaged, Life, Ammo)	OpenFIRE wiki: GunRecoil “doesn’t work well” in Operation G.H.O.S.T.
Time Crisis 5	Namco ES3X	TeknoParrot + DemulShooterX64	Yes	DemulShooter: Yes
Point Blank X	Namco ES4	TeknoParrot + DemulShooter	Yes	DemulShooter: Yes
Time Crisis 4	Namco System Super 256	TeknoParrot (embedded PCSX2)	Per TeknoParrot profile	Not in DemulShooter’s list
Aliens Armageddon, Terminator Salvation, The Walking Dead, Jurassic Park (2015), Halo: Fireteam Raven	Raw Thrills PC	TeknoParrot + DemulShooter	Yes (Halo: four)	DemulShooter: Yes
Luigi’s Mansion Arcade	Sega Nu	DemulShooterX64 (seganu)	Yes	DemulShooter: Yes
Silent Hill: The Arcade	Taito Type X	DemulShooter (ttx)	Yes	DemulShooter: custom Yes (no Damaged, Life, Ammo)	Keyboard keys must be set
Time Crisis 2, 3, Crisis Zone, Vampire Night	PlayStation 2 (GunCon 2)	PCSX2	Second gun only in gamepad mode	None; gun’s own	Calibration Shot bind
PlayStation GunCon and Justifier games	PlayStation	DuckStation (Raw Input) or Beetle PSX	Yes	None; gun’s own
Duck Hunt and Zapper games	NES	RetroArch, FCEUmm	Yes (two ports)	None; gun’s own
Super Scope and Justifier games	SNES	RetroArch, Snes9x	Yes	None; gun’s own
Menacer, Justifier and Light Phaser games	Mega Drive, Master System	RetroArch, Genesis Plus GX	Yes	None; gun’s own
Operation Wolf Returns: First Mission	Windows (Steam)	DemulShooterX64 + Unity plugin	Yes	DemulShooter: Yes	Non-VR version only
The House of the Dead: Remake	Windows (Steam)	Arcade Plugin + DemulShooterX64	Yes	DemulShooter: Yes	Original release: no light guns
The House of the Dead II, III, Overkill (PC)	Windows	DemulShooter (windows)	Yes	DemulShooter: Yes
Heavy Fire: Afghanistan	Windows	DemulShooter	Up to four	DemulShooter: Yes	Cover by aiming at a screen edge

6.8 Proposed decisions

For the category’s decision log, each with its reason; none is an owner decision yet.

Table 6 — 6.8 Proposed decisions

#	Area	Proposal	Why, and why not the alternatives
PW-1	Hook program	QMamehook by default; OutputHooker trialled for Windows-message sources; MAMEHooker as fallback	No administrator rights and OpenFIRE-aware vs MAMEHooker’s admin requirement and COM1–4 limit (Volume 5)
PW-2	MAME configuration	rawinput, lightgun_device lightgun, mapdevice on VID_F143&PID_000n	Per-gun input and stable P1/P2; win32 is one gun (Volume 2)
PW-3	Game-driven recoil commands	Pulse on the output’s “on” state (`F0x0	F0x2x1form), neverF0x1` holds, for every game
PW-4	Bench display	A VRR-capable monitor as the reference display, with the TV’s game mode measured separately	One to two frames separate the presentation modes (6.1)
PW-5	Second-gun fallback	Gamepad mode only where a program has no Raw Input (PCSX2)	Raw Input is per gun and absolute everywhere else (Volume 1)

6.9 What could not be verified

• Every latency figure in 6.1 beyond the gun’s own share is a frame-period model, not a measurement.
• Whether each program named in 6.2 actually runs in Independent Flip on a given PC (check with PresentMon).
• Flycast’s, DuckStation’s and PCSX2’s own low-latency option names (6.3).
• Every row of 6.7 is a documented route, not a tested one; the recoil column is DemulShooter’s or MAME’s published capability, not observed on this gun.

Sources

1. Microsoft, “For best performance, use DXGI flip model”, Microsoft Learn, updated 15 April 2026, https://learn.microsoft.com/en-us/windows/win32/direct3ddxgi/for-best-performance—use-dxgi-flip-model — read 2026-10-20.
2. MAMEdev, MAME 0.289 documentation, lowlatency, waitvsync, syncrefresh, devicemap — read 2026-10-20.
3. libretro, RetroArch v1.22.2 intl/msg_hash_us.h (latency settings) and the libretro “Latency” and “Run Ahead” guides — read 2026-10-20.
4. argonlefou, DemulShooter wiki v17.10, “Outputs” (extracted per game to research-demulshooter-outputs.txt), “Usage”, “Model2”, “Demul”, “Windows games” — read 2026-10-20.
5. teknogods, TeknoParrotUI game profiles and metadata — read 2026-10-20.
6. The Supermodel project, Docs/README.txt, Src/OSD/Outputs.cpp; flyinghead, Flycast source; PCSX2 and DuckStation source — read 2026-10-20.
7. MAME driver sources for output names, as listed in Volume 2 — read 2026-10-20.
8. TeamOpenFIRE, MAMEHOOKER Documentation — read 2026-10-20.
9. This category’s own volumes for the gun’s latency share: Aiming Maths and Calibration, Volume 7; Fully Wired, Volume 5.