NOBD-ZERO
The most over-engineered fightstick PCB ever designed.
Four chips into one. One controller.
A flagship fightstick PCB on a single chip: dual-core RISC-V, open firmware, native retro consoles, up to 16 kHz USB target with USB 3.0 headroom, designed to outspec the category. It also fixes a real bug. The classics ran in sync with their hardware, frame by frame. On modern hardware they fall out of sync: fast polling and the game’s once-per-frame read split your two-button intent across a frame boundary, dropping an input. A tunable sync window puts both back on the same frame.
The Founding 100 lock $150 · $199 after
The Founding 100 lock $150 · retail $199 · no payment now · non-binding
- 16 kHz
- USB-HS · target
- ~40 kHz
- LAN mode · target ↗
- One chip
- Dual-core RISC-V
- Native
- Retro
drag to rotate · NOBD-ZERO pre-prototype
Watch the drop. Then watch it land.
Same stick, same two-button input. With sync off, the second press is dropped and the move never comes out. Turn sync on and both land. This is the open NOBD fix running live, not a render.
Captured on a real fightstick · Marvel vs Capcom 2 · tap to enlarge
You pressed two buttons. A jab came out.
That is a real, measurable bug. On the classics, run on modern hardware, 1000 Hz polling can split two near-simultaneous presses across a frame boundary, so one input drops and the move you meant never comes. We found it, measured it, and removed it with a tunable sync window.
The only variable left is you.
No human presses two buttons at the same instant.
In our testing, two fingers land 2 to 8 ms apart. That gap is human, not sloppiness. A 1000 Hz stick reports it one millisecond at a time, so a 3 ms split goes out as three reports with only the first button down, before your hand has finished the press.
Why it matters
Think of a camera that snaps one photo per frame. If your two presses land on either side of the shutter, the photo only catches the first one.
One game frame · 16.67 ms · one read, by design
Your whole input should land inside one frame. On synced hardware, it did.
Zoom in · USB reports at 1000 Hz · 1 ms each
If the frame's one read lands in those first 3 ms, the game sees LP only and your move dies. The original hardware read you once per frame, so your whole input had a full 16 ms to land together. Sub-ms polling reports your intent before it has finished forming.
A decades-old game, running on hardware it was never built for.
These games were built for arcade boards and consoles, where everything moved to one clock. The screen refreshed at 60 Hz, the game ran at 60 frames a second, and the controller was read once per frame, all locked into the same beat. That is why old consoles felt so consistent. Your two buttons, a few milliseconds apart, got swept into the same read and came out together. It just worked. You never had to think about it.
Then the games left that hardware. On a modern PC, emulator, or Fightcade, your stick and the game no longer share one clock, and your inputs take a longer, looser path before the game looks. Those same two buttons can land on opposite sides of a frame, and the game, still reading once per frame like it always has, sees only the first. You get one button instead of the move you meant. Newer fighters quietly added forgiveness in software, but these classics are frozen in time and can't be patched. The only place left to put the sync back is the controller.
This is a retro, Fightcade, and emulation problem. Modern fighters added leniency; we don't claim they drop your inputs.
Read the fix
One open-source function you can read
The exact sync logic lives in our public fork, not a slide. The actual code.
src/gp2040.cppIndependently measured
Slower boards group presses better
A relay-rig test (Vodka's, not ours) found older and slower boards register simultaneous presses more consistently than fast modern ones. Outside evidence, pointing the same direction we are.
Read Vodka’s testTry it yourself
Run it free on your PC
NOBD Desktop brings the same sync window to the Steam version of Marvel vs Capcom 2, in software. No NOBD board, no flashing, it just hooks the game and uses the stick you already own. Feel the fix before you spend a cent.
NOBD DesktopI'm not a firmware engineer. I'm a cloud support engineer, two decades in IT and infrastructure, and a competitor who came back to Marvel vs Capcom 2 after 15 years.
Truth is, I walked away from this game years ago, not long after it got ported around. It never felt the same to me. The timing was off in a way I couldn't put my finger on, so I did what everybody does. I blamed myself and stopped playing.
Years later I fired it up again on PC, and the dashes kept dying. My first thought was, great, I'm getting old. But come on. Anyone can dash. I am not THAT old. Is it me? I kept asking that until the troubleshooter in me took over. Two decades in IT and infrastructure teach you one thing better than anything else: how to chase a problem. Gather the evidence, isolate the symptom, follow the leads until the picture is clear. And the picture here looked exactly like a desync between two clocks. The stick and the game were out of step, and fast polling was exposing the gap. It was not me. That is where the research began.
So I used everything I had: decades of troubleshooting instinct, modern AI tools, and an obsession that wouldn't quit. I open-sourced the fix, showed every line, and set out to build the most over-engineered competitive PCB anyone has tried to ship, so you never have to take my word for any of it.
For the perfectionist. For the grinder. The only variable left is you. Your training finally pays off.
NOBD · cloud support engineer, competitor
Every edge we could engineer, in one board.
One chip does the work of four. Up to 16 kHz USB with USB 3.0 headroom. Native retro consoles. Ethernet built into the silicon. Every input mode, Hall triggers, firmware you can read and fork. No shipping product combines them, and the open board everyone else uses is capped at 1000 Hz by its own silicon. We build in public, and every claim here is checkable.
Specs below are targets. Pre-prototype hardware.
Up to 16 kHz USB · USB 3.0 on board
8 kHz standard, 16 kHz with the NOBD companion: every input reported in 0.0625 ms, 16x the 1000 Hz standard. And the chip carries a full USB 3.0 SuperSpeed engine on die, headroom no fightstick has ever shipped.
One chip · dual-core RISC-V
The last board needed four chips. This one needs one. A single WCH CH32H417 with two RISC-V cores does the work of a whole board: one core keeps your inputs clean, dedicated hardware engines handle every poll, so timing never gets blocked by USB or network work.
No CPU in the hot path
Speed you can feel is really consistency. The button-to-console transaction runs on dedicated silicon with a pre-computed response, not a software loop, so your timing does not drift when the chip gets busy. Same input, same path, every frame. Low jitter is the whole point.
Ethernet on-chip · LAN mode
The first fightstick PCB built around a real network stack, and now it lives inside the chip itself, no separate network IC. A companion app presents your stick as a virtual controller fed over Ethernet, bypassing USB polling entirely. Not faster on average. Deterministic. Your input reaches the game the moment it changes, the same way every time. Have a spare Ethernet port on your PC? Direct connect is supported too, no crossover cable required.
Target: your input hits the game in ~25 µs. A standard 1000 Hz stick takes ~1,000 µs, so your press can reach the game nearly a full millisecond sooner, the equivalent of about 40,000 Hz polling. Ethernet on the die removes the ~12 µs SPI hop the old external network chip added, so the floor gets lower still.* (*Theoretical, pre-prototype, not yet measured. Real numbers go to inputlag.science before we claim them.)
Native retro · Brook-cable ready
Dreamcast is native over Maple Bus, no adapter and no dongle, and it is running in the open NOBD firmware today. On the Zero board it runs on the chip’s PIOC, a built-in protocol engine, so the retro bus takes zero main-CPU. The retro jack uses the standard Brook retro-cable pinout, so the Brook adapter cables you already own plug right in. More classics roll out in firmware.
Every input mode, however you play
One board, every system: output modes for Xbox, PlayStation 4, Switch, PS3, the Original Xbox, and PC DirectInput, with per-game profiles you switch on the fly. Optional Hall-effect switches let you set the actuation point in software.
On-chip crypto · saves & replays
The chip has real hardware encryption built in. On the roadmap: encrypted VMU saves and encrypted tournament replays, done in silicon, not bolted on in software.
Open firmware + Brook 20-pin
Runs the fully open NOBD firmware. Read every line, fork it, and drop it into your existing stick via the standard Brook 20-pin.
Two modes · Sync & Raw
Sync mode groups your presses so they land together, on a window you can tune (5 ms by default), trading a sliver of speed for rock-solid consistency. Raw mode runs full speed at minimum latency. Fast when you want it, synced when you need it.
Faster, drawn to scale.
Bar = speed vs 1000 Hz wall
Our target. Over Ethernet, event-driven.
Our target with the NOBD companion. 16 kHz, same board.
Our target out of the box. 8 kHz over USB High-Speed.
The fastest fightstick ever measured. Right on the wall.
The firmware we forked. Stuck at the 1000 Hz USB cap.
A stock PS4 pad.
Slowest on the board.
Every grey bar is a real measurement from inputlag.science , drawn as speed relative to the 1000 Hz wall (1 ms = 1×). The fastest fightstick ever tested (Brook UFB) sits right on that wall, and the open RP2040 boards GP2040-CE runs on are capped there too. Our targets aim 8× past it on USB out of the box, 16× with the companion, and 40× on LAN.
* NOBD figures are design targets on pre-prototype hardware, not yet measured. GP2040 is the 1000 Hz USB cap, not an inputlag.science entry. We will submit ours before we claim a number.
No single product combines all of this. And we will submit our latency for independent testing, because in this scene you don't trust specs, you test them.
The board we're building. Renders today, silicon next.
NOBD software
The open-source fix, already running on GP2040-CE boards.
The board
The one-chip CH32H417 design: up to 16 kHz USB, USB 3.0, Ethernet on-chip. In design.
LAN Mode + more
LAN Mode, gigabit tier, more retro consoles, encrypted netplay. Planned.
Receipts, not trust.
The part competitors won't do. Every claim on this page is checkable, so don't believe us. Verify us.
- 01
Read the fix
One open-source function, the actual sync logic, in our public fork. Not a slide.
src/gp2040.cpp - 02
Measure your own gap
Don’t believe 2–8 ms? Measure yours with the open Finger Gap Tester.
Finger Gap Tester - 03
Try it free on your PC
Don’t believe any of it? NOBD Desktop drops the same sync window onto the Steam version of Marvel vs Capcom 2, in software, using the stick you already own. No NOBD board, no flashing, no money down. Feel the difference yourself. (MvC2 today; more games coming.)
NOBD Desktop - 04
Independently measured
A relay-rig test (not ours) found older and slower boards register simultaneous inputs more consistently. Outside evidence, not our marketing.
Read Vodka’s test
You drilled the same input for years, never sure if the drop was you or the board. We removed the doubt. The sync window groups your inputs onto one frame, so what comes out is what you did. The only variable left is you. Own every result.
And the fix is one feature on the most over-engineered fightstick PCB anyone has tried to ship. Read the code. Measure your gap. Verify our latency yourself. We are not asking for your trust. We are handing you the receipts.
Remove the variable. The rest is on you.
Reserve yoursThe Founding 100 lock $150 · retail $199 · no payment now
Build it with us.
NOBD is built in public, with the people who actually play these games. Jump into the Discord, ask questions, try the software, and help decide where this board goes next. This is your scene too.
Join the DiscordTalk to the bots
An MvC2 Oracle for frame data and matchup questions, plus a bot that drops your feature requests and bug reports straight into the backlog.
Shape the firmware
Ask for a feature, report a bug, and watch it get built. It's all open source. The community speaks; the firmware follows.
Try it, then talk
Run the free NOBD software on a setup you already own, feel the fix, then come tell us what you found. Not a hype channel. A workshop.
Lock your price. Own every result.
The first 100 reservations lock the $150 founding price and a founding serial. Retail lands at $199. No payment now, nothing binding, and we tell you the final price before you ever owe a cent. Reservations are how we decide how many to build.
The Founding 100 lock $150 · $199 after
One email. We tell you the final price before you owe a cent. No spam, unsubscribe anytime.