Windows 路线:在 Windows 启动之前执行的 EFI 载荷

本站记录的三项上游项目都是 Linux 内核模块。Windows-GPU-UNLOCK 是另一套实现:它单独建立一个 EFI 系统分区,把寄存器操作放进一个交给 Windows Boot Manager 的载荷里执行。该项目也没有公布同一张卡的原厂/解锁对照,所以这一页给的是已归档配置与启动路径,而不是提升倍数。

这不是 Linux 模块,也不能替代它

这里的内容不适用于 Linux 安装,内核模块也不适用于 Windows。两者共享同一个思路——在 NVIDIA 驱动绑定之前写入速率选择寄存器——但启动路径、工具链和失败模式都不同。如果你在 Linux 上,请使用本站的显卡列表与实测数据。

截至目前,这个发布是什么

最新标签是 v0.3.1,是一个只提供源码的预发布版本。在假定某张卡受支持之前,请先读下面的范围说明。

已归档与已上报的硬件

一个配置只覆盖某个特定系统里的某一张卡。同一型号的 GPU 换一块主板就不再被该配置覆盖,并且只有第一行是维护者自己在测试中归档的。

显卡核心状态路径
RTX 3090 FE 24 GBGA102维护者已归档Internal headless — default profile (ASUS TUF B650M)
RTX 3070 Ti 8 GBGA104已上报硬件成功Internal headless (ASUS TUF B660M-E D4)
RTX 3090 Ti 24 GBGA102已上报硬件成功Headless USB only (Gigabyte X870 Aorus Elite WiFi7)
RTX 3060 12 GBGA104已上报硬件成功Headless USB only (ASUS Prime B660-Plus D4)
RTX 3080GA102仅上游目标No dedicated profile in this port

四条启动路径,彼此不能互换

走哪条路径取决于显示是否留在独立显卡上,以及你能否使用 U 盘。每条路径在仓库里都有各自的指南。

路径显示与输入指南
Internal headlessMonitors on the RTX card, display temporarily removed, no T/U/Enter input打开指南
Headless USBProfile-specific wrapper and child staged on a FAT32 USB stick打开指南
Internal AMD-display manual / autoMotherboard or iGPU display preserved; the manual variant needs U打开指南
Legacy USB manualAMD/iGPU display profile; U once, then Boot Override to Windows without a reset打开指南

开始前需要满足的条件

真正重要的顺序

1. Prepare a second ESP, not a data partition

The port boots from its own EFI System Partition and leaves Windows' ESP and recovery partition untouched. It does not create, shrink, delete or format anything: you prepare a GPT FAT32 partition of at least 1 GiB and give it a letter.

An ordinary FAT32 data partition is not a substitute. The tools also do not disable BitLocker — decrypt C: and turn protection off first.

2. Configure and build from an elevated PowerShell

Both scripts run from the repository root in an administrator PowerShell. Configuration and build stage no boot entry by themselves.

.\internal-boot\Configure.ps1 -UnlockDriveLetter U `
  -WindowsEspGuid 'YOUR-WINDOWS-ESP-GUID'
.research\headless\build-unlock.ps1 `
  -Clang 'C:\TOOLS\LLVM20\bin\clang.exe' `
  -LldLink 'C:\TOOLS\LLVM20\bin\lld-link.exe'

Use your own partition GUID and tool paths. Without an action switch, `Headless.ps1 -Assets '.\assets'` only stages files.

3. Register a separate F8 test entry first

Registering adds its own test entry and keeps your existing boot priority. Automatic boot is a separate, explicit action and stays off until the F8 test has actually worked.

.\internal-boot\Headless.ps1 -Assets '.\assets' -Register

On the tested ASUS firmware the entry appears as F8 → GPU Headless UNLOCK - EXPERIMENTAL.

4. Run one F8 test and wait

Save your work and fully power off, then pick the test entry. Do not press T, U or Enter. The screen goes black while the RTX GOP is removed; the child attempts once and the wrapper starts Windows directly.

Do not reset between the unlock and Windows. The two Esc windows differ: Esc in the initial wrapper boots Windows without the release, while Esc in the later three-second child window cancels into a stop and does not restore the display.

5. Only then enable automatic boot

After the F8 test succeeds and the display recovers, automatic boot can be enabled. That action places headless first, Windows Boot Manager second, and disables Windows Fast Startup. Registering the test entry alone does nothing like this.

.\internal-boot\Headless.ps1 -EnableAutomatic -ConfirmedSuccess

`-ConfirmedSuccess` is the acknowledgement that you personally saw the test boot work.

6. Recovery is Windows-first and independent of the build

If the handoff fails or Windows does not return, power off fully and choose F8 → Windows Boot Manager, or run the restore switch. Windows-first recovery works without saved state.

.\internal-boot\Headless.ps1 -RestoreWindowsFirst

Restoring the previous Fast Startup setting does need the saved state, so keep it and keep backups.

那次归档测试测到了什么

原始 B650M / RTX 3090 FE 测试给出了这些 CUDA 12.6 / cuBLAS 12.6、4096 x 4096 下的数值,取三次运行的中位数。这是解锁后显卡的绝对值——项目没有公布对应的原厂数值,因此这不是提升数据。

项目数值
FP32 pedantic22.85 TFLOPS
FP16 Tensor / FP32 accumulate124.97 TFLOPS
BF16 Tensor / FP32 accumulate128.93 TFLOPS

那次归档测试留下了什么状态

成功的那一次子载荷在解锁后上报了如下状态,这是显卡被干净地交给驱动的证据。

许可证

GPL-2.0-only for the port and the reused core; EDK2 and separately attributed components keep their own licences. Based on rough2515-hub/rtx3080-unlock-public and Zagitta/gpu-unlock. NVIDIA firmware, PROM/BIOS images, drivers and CUDA DLLs are not redistributed in the source release.

GitHub 上的 impacifist/Windows-GPU-UNLOCK

常见问题