One small app that controls your Lian Li fans, AIO pump, RGB lighting and TL LCD screens on Windows. No L-Connect. No iCUE. No GIGABYTE Control Center. No FanControl. No background services from four different vendors fighting over the same hardware.
That is the entire application — everything it does is on that one screen.
Status: works, used daily, but built around one specific machine. Read Supported hardware before you try it — this is not a universal RGB tool.
A typical build ends up needing a separate resident app per vendor: one for the fans, one for the AIO, one for the RAM, one for the motherboard. Each one autostarts, each one runs a background service, and together they cost hundreds of megabytes of RAM to change a colour twice a year.
This replaces all of them with a single PowerShell app that talks to the hardware directly — through Windows' built-in HID stack and one signed, sandboxed kernel driver (PawnIO).
Linux users already have an excellent answer to this problem
(sgtaziz/lian-li-linux). Windows
users did not. That's the gap this fills.
| Capability | Hardware | Admin needed |
|---|---|---|
| Fan speed (all fans on one master slider) | Lian Li UNI FAN TL | No |
| Pump speed | Galahad II Trinity AIO | No |
| Fan RGB | UNI FAN TL | No |
| Pump head RGB (inner / outer rings separately) | Galahad II | No |
| AIO fan-channel RGB | Galahad II | No |
| RAM RGB | Corsair Vengeance DDR5 | Yes |
| LCD screens — static images and animated GIFs, per fan | UNI FAN TL LCD | No |
| Read CPU temperature and package power | AMD Ryzen | Yes |
| Read GPU temperature, load, clocks, fan RPM | NVIDIA | No |
| Read fan and pump RPM | Lian Li | No |
| Start with Windows (optional, two separate toggles) | — | — |
If you decline the admin prompt, everything in the "No" rows still works.
- GPU fan control. Left on NVIDIA's automatic curve on purpose. Pinning a GPU fan low is a thermal risk and the gain is minimal.
- Motherboard header fans. These run off the board's ITE Super I/O chip. On
the test machine that chip enumerates as
048D:57DB"ITE Upgrade Mode" — a firmware-update endpoint. A wrong write there can permanently brick the RGB controller or the EC, so this project never writes to it. Use your BIOS fan curve for header fans; it's the right place for them anyway. - Anything that permanently rewrites device state. See Safety.
Everything below is verified working on the development machine:
| Device | USB ID |
|---|---|
| Lian Li UNI FAN TL controller | 0416:7372 |
| Lian Li Galahad II Trinity AIO | 0416:7373 |
| Lian Li UNI FAN TL LCD screens (×3) | 04FC:7393 |
| Corsair Vengeance RGB DDR5 (×4 modules) | SMBus 0x18–0x1B |
| AMD Ryzen 7 9700X (Zen 5, Family 1Ah) | via PawnIO |
| NVIDIA RTX 4070 SUPER | via NVAPI |
-
More TL fans on the same controller. The controller reports its own fan inventory over a handshake, and the app drives whatever comes back. Nothing assumes a fan count. Protocol allows 4 ports × 16 fans.
-
More than one TL controller. All matching controllers are enumerated and opened; each fan remembers which controller it belongs to, so the master slider and the colour picker reach every fan on every controller. If one controller fails to open, the others still work. Only ever tested with a single controller, so this is "should work", not "verified".
-
Other AMD Ryzen CPUs (Family 17h/19h/1Ah). The code probes candidate SMN addresses and picks the one returning plausible values.
-
Other NVIDIA GPUs.
-
A different motherboard, for RAM RGB. The SMBus port the DIMMs sit on varies by chipset, so it is probed at startup across all five ports rather than assumed. Colour is then written to the whole
0x18–0x1Fcontroller range, never to a narrowed "detected" list — see the ACK note under Protocol notes for why that distinction matters. -
More or fewer than 4 DIMMs. All eight possible slot addresses are written; writes to empty addresses are harmless no-ops.
-
Any number of LCD screens. Rows are built from the screens actually present, and the window grows or shrinks to fit. With exactly three screens the rows are labelled top/middle/bottom, since that physical mapping was measured; with any other count they are labelled by index, because the physical order is not known.
- Intel CPUs, AMD GPUs, non-Corsair RAM, other brands of RGB fan. Not implemented.
- Windows 10 or 11
- Windows PowerShell 5.1 (ships with Windows — no install needed)
- PawnIO — only for RAM RGB and CPU temperature:
winget install --id namazso.PawnIO -e - Administrator rights, only for those same two features
- Download the latest release
and extract it somewhere permanent — not your Downloads folder, because
the "start with Windows" option stores this path. (
git cloneworks too.) - Unblock the extracted files. Windows marks everything that came out of a
downloaded zip, and clicking through forty files one at a time is not
reasonable. Open PowerShell in the folder you extracted to and run:
Get-ChildItem -Recurse | Unblock-File
- Install PawnIO if you want RAM RGB and CPU temperature.
- Double-click
Start.cmd.
Windows will warn you the first time. This is expected and it is not a statement about the code — it means the files are unsigned and downloaded from the internet. Code signing certificates cost money and this project has none.
Click More info → Run anyway. The Unblock-File line in step 2 above
clears the same mark on every file at once.
If that trade is not acceptable to you, that is a completely reasonable position.
Everything here is plain text: read CoolApp.ps1 and the lib/ folder before
running anything. That transparency is a deliberate reason this is shipped as
source rather than as a compiled binary.
Some antivirus products may also flag this. The program legitimately loads a kernel driver, writes to the SMBus, runs elevated and can register itself to start with Windows — behaviour that heuristics associate with malware. Judge it by reading the source.
Everything is in one window:
- Speed slider — moves every fan and the pump together. The pump has a hard floor of 40% and never follows the fans below it.
- Colour — pick a colour, press Apply. Fans, pump head, AIO fan channel and RAM all change together.
- Fan screens — assign an image or GIF per fan, set brightness, press Apply to screens.
- Start with Windows — two independent checkboxes, both off by default. With This app ticked, the app starts straight into the notification area at logon; no window opens over whatever you were about to do.
- Minimise to the notification area — minimising the window hides it and leaves an icon next to the clock. Double-click that icon, or pick Show window from its right-click menu, to bring the window back. Closing the window still quits the app. Sensor polling pauses while the window is hidden, so nothing is read from the devices for a window nobody is looking at.
- Language — English and Turkish, bottom right. The default follows your
Windows display language; change it and the window switches immediately,
without restarting or losing your settings. The choice is remembered in
dil.txt.
Exception messages coming from the device libraries are English in both languages — they are diagnostics, and the technical language of this project is English.
Supported formats: gif jpg png bmp webp tif. Images are centre-cropped
and scaled to 400×400.
Animated GIFs are streamed, not stored. The screens have no memory for
animation — each frame is sent over USB as a JPEG. That's why animation needs a
running process, and why LcdDaemon.ps1 is a separate background process: so
the screens keep animating after you close the app. Static images persist on the
device on their own and need no daemon.
Measured throughput across three screens (PowerShell 5.1):
| JPEG frame size | Frame rate |
|---|---|
| ~14 KB | ~17 fps |
| ~19 KB | ~14 fps |
| ~27 KB | ~10 fps |
Frame rate is bound by bytes per frame, not frame count. Quality 55 is the sweet spot between sharpness and smoothness; that's the default. The hardware caps at 30 fps and a single image cannot exceed 65535 bytes.
To stop the screens, use Stop screens in the app, or:
powershell -NoProfile -ExecutionPolicy Bypass -File LcdDaemon.ps1 -StopThis project touches a kernel driver and the system SMBus. Several guards are built in on purpose, and they are not configurable:
- SMBus writes to
0x48–0x4F(DDR5 PMIC) and0x50–0x57(DDR5 SPD hub) are permanently blocked. A wrong write to a DDR5 SPD hub or PMIC can destroy a memory module. RGB controllers live at different addresses. This block cannot be disabled by a parameter. - Every SMBus operation runs under a named mutex. If the lock cannot be acquired, the operation does not run. It never proceeds unsynchronised.
- The pump never drops below 40%, no matter where the master slider goes.
- The motherboard ITE chip is never written to.
- CPU access is read-only.
ioctl_write_msris deliberately not wrapped — writing MSR/SMN registers can destabilise the CPU or corrupt persistent settings. - LCD commands that permanently alter device state are deliberately not
implemented:
63(write serial),0x45(AVI),0x47/0x48(boot images). A bad write there could leave a screen unusable at boot.
Nothing here is a substitute for reading the code yourself.
No warranty. See License. You run this at your own risk.
Possibly the most reusable part of this repository. All of it is verified against real hardware.
Report ID 0x01, 64-byte packets, 6-byte header. Handshake 0xA1 returns the
controller's own fan inventory: 3 bytes per fan (port/index/detected bits,
then RPM big-endian).
Duty is documented elsewhere as 0–255 PWM, but measured behaviour on
TL_Series_ControllerV0.62 is a 0–100 scale — every value above 100 yields
the same maximum RPM.
Same packet family as the TL controller but different command bytes: 0x81
handshake, 0x8A pump PWM, 0x8B fan PWM, 0x83/0x85 lighting. This device
has no coolant temperature sensor.
HID output report ID 0x02, exactly 512 bytes, 11-byte header, all multi-byte
fields big-endian:
[0] report id (0x02)
[1] command
[2..5] total data size (4 bytes, BE)
[6..8] packet number (3 bytes, BE)
[9..10] this packet's payload length (2 bytes, BE)
[11..] payload (max 501 bytes)
Screen is 400×400, JPEG only. Command 0x41 writes a single image and ACKs every
packet; 0x46 streams frames without ACK.
The trap that cost the most time: do not switch the device to mode 4
(ShowAppSync) before streaming. It sounds like the correct mode and it is not —
the device does not render 0x46 frames there and you get a frozen screen with
no error. Stay in mode 1 (ShowJpg).
Colour is one 32-byte block write to register 0x31:
[led count][R,G,B] × 10 [CRC-8], CRC-8 poly 0x07, init 0x00, no reflection,
no final XOR. RGB controller address = SPD address − 0x38.
The older DDR4-era protocol (0xA4/0xA5/0xA6 mode registers, 0xB0–0xB2
separate R/G/B bytes) has no effect on DDR5 modules. It was tried; nothing
changed.
ACK probing is unreliable on these controllers. Addresses 0x19 and 0x1B
answered a QUICK probe only sometimes — but block writes worked every time,
visually confirmed. Scanning for devices silently dropped half the modules. Use a
fixed address list, not a scan result.
Read from the module source, not guessed:
addresses[] = [0x0B00, 0x0B20]
port_to_reg[] = [0b00, 0b00, 0b01, 0b10, 0b11]
reg_to_port[] = [0, 2, 3, 4]
Valid port range is 0–4 — five ports, not four. Port 0 is primary base / mux
0, port 1 is the auxiliary base 0x0B20, and ports 2–4 are the primary base at
mux 1/2/3. Scanning only 0–3 will miss devices; that mistake is why two RAM
modules initially looked unreachable.
CoolApp.ps1 the application (speed, colour, screens, startup, status)
LcdDaemon.ps1 independent background process that feeds the LCD screens
Start.cmd double-click this to run the app
assets/ images used by this README
lib/ device libraries
Dil.ps1 UI text in English and Turkish
HidCore.ps1 HID enumeration and I/O (embedded C# via Add-Type)
PawnIO.ps1 kernel driver interface
Smbus.ps1 SMBus access + the permanent write guards
LianLiTL.ps1 UNI FAN TL
LianLiGA2.ps1 Galahad II AIO
LianLiLcd.ps1 TL LCD screens
CorsairDram.ps1 Corsair DDR5 RGB
AmdCpu.ps1 AMD temperature / power (read-only by policy)
NvApi.ps1 NVIDIA sensors
modules/ PawnIO kernel modules (LGPL 2.1, see modules/COPYING)
tools/ standalone diagnostics and scanners
The tools/ folder holds the probes and scanners used to work the protocols out.
They are read-only unless their name says otherwise, and they are useful when
porting to different hardware.
This project would not exist without prior reverse-engineering work by others. The protocol knowledge here is derived from these projects, not independently discovered:
sgtaziz/lian-li-linux(MIT) — the TL LCD protocol comes fromcrates/lianli-devices/src/tl_lcd.rs. A far broader project than this one; if you are on Linux, or you own Lian Li hardware this app doesn't cover, go there first.- OpenRGB (GPL-2.0) — the
Corsair DDR5 register map and packet format come from
CorsairDRAMController. - PawnIO by namazso — the sandboxed kernel driver, and
the
.binmodules redistributed inmodules/(LGPL 2.1). - Open-source reverse-engineering work on the Lian Li HID protocols generally.
If you want broad hardware support, contribute to OpenRGB — it will reach far more people than this repository will.
Built and maintained by CihanStudio — cihanstudio.com.
CihanStudio builds AI assistants and digital-transformation tooling for businesses. This project is a hardware side-project rather than commercial work, published in the open because the protocol notes above are worth more shared than sitting on one machine.
Issues and pull requests are welcome, particularly hardware reports from configurations different to the one this was developed on.
GPL-2.0. See LICENSE.
GPL-2.0 was chosen deliberately: this project uses protocol knowledge derived from OpenRGB, which is GPL-2.0. Matching that licence removes any ambiguity about compatibility and keeps the work in the same commons it came from.
The PawnIO modules in modules/ are LGPL 2.1 and carry their own licence text in
modules/COPYING.