← Back to AI Insights
Gemini Executive Synthesis

The `nv-uefi-unlocker` tool, specifically its multi-GPU detection and unlocking functionality within a UEFI environment, leading to incorrect GPU enumeration in the OS.

Technical Positioning
Reliable, comprehensive NVIDIA GPU unlocking across multiple GPUs and diverse EFI-bootable systems, ensuring all detected hardware is correctly exposed and functional within the operating system.
SaaS Insight & Market Implications
This issue reveals a critical failure in the `nv-uefi-unlocker`'s core promise: enabling full GPU functionality. Despite successfully detecting two GPUs and completing its 'multi-GPU pass' during the UEFI pre-boot phase, only one GPU is visible in Windows 10. This indicates a fundamental breakdown in the unlocker's ability to persist its modifications or correctly hand off control to the OS for all targeted hardware. For a specialized utility like this, partial success is a complete failure from a user perspective, especially in multi-GPU configurations common in high-performance computing or professional workstations. The `ACCESS_DENIED` error for `tx-select-bdf.txt` might be a symptom, suggesting permission or configuration issues preventing full application of the unlock. This directly impacts the tool's value proposition and user trust.
Proprietary Technical Taxonomy
nv-uefi-unlocker UEFI pre-driver NVIDIA GPU unlocker Windows 10 dual RTX 3060 (GA106) EFI-bootable system loader::pci: ENUMERATE NVIDIA display adapters

Raw Developer Origin & Technical Request

Source Icon GitHub Issue Nov 4, 2026
Repo: physicsdolphin/nv-uefi-unlocker
When running 2.1.10 on a dual RTX 3060 (GA106) box, only 1 GPU shows up in Windows 10

### 工具版本 / Tool version

2.1.10

### 问题组件 / Affected component

系统内状态或性能 / In-OS state or performance

### 实际与预期行为 / Actual and expected behavior

Actual:
Two GPUs are found:
[ 6590ms] INFO : loader::pci: ENUMERATE: 2 NVIDIA display adapters present:
[ 6619ms] INFO : loader::pci: ENUMERATE: candidate 0000:03:00.0 dev=10de:2504
[ 6649ms] INFO : loader::pci: ENUMERATE: candidate 0000:04:00.0 dev=10de:2504
...
[ 8818ms] INFO : loader::tx: TX-ASSET-DIAG: \usbroot\tx-select-bdf.txt fs open failed ACCESS_DENIED
[ 8848ms] INFO : loader::tx: TX-ASSET-DIAG: \tx-select-bdf.txt absent
[ 8884ms] INFO : loader::pci: ENUMERATE: no tx-select-bdf marker - every supported card runs, in enumeration order (2 found)
...
[ 40593ms] INFO : loader::tx: FINAL: transaction complete, all DMA released
[ 40606ms] INFO : loader: SELECTOR RESULT: transaction complete
[ 40620ms] INFO : loader::tx: HANDOFF-OUT: ss0=0x88888888 ss1=0x8 readout1=0x00000000 feat_plm=0xffffffff wpr2_hi=0x0 pmcsr=0x0008 dstate=D0 | tu10x: plm=0xbadf5040 ss0=0xbadf5040 ss1=0xbadf5040 readout=0xbadf5040 | tu10x:unlock=false ga/ada:ss0_unlock=true ss1_unlock=true plm_open=true
[ 40634ms] INFO : loader: TX: multi-GPU pass complete (2 target(s): 2 attempted, 0 skipped - per-card verdicts above)
[ 40635ms] INFO : loader::chainload: PHASE: CHAIN
[ 40646ms] INFO : loader::chainload: CHAIN-PRE: start-time PE header probe ok - StartImage the preloaded '\EFI\Microsoft\Boot\bootmgfw.efi'

All appears to go well, but wh...

Developer Debate & Comments

physicsdolphin • Nov 4, 2026
Thanks for the log — it pins the problem down sharply, and the conclusion runs against intuition: **the UEFI stage did not "lose a card". Both cards ran the transaction to completion. The problem happens after the handoff back into Windows.** ## What the log already establishes - `[6590ms] ENUMERATE: 2 NVIDIA display adapters present` → both `03:00.0` and `04:00.0` are present (`10de:2504`); - Both transaction passes reach the end: `target 1/2` → `FINAL`, `target 2/2` → `FINAL`, `2 attempted, 0 skipped`; - The POST-FLR read-back passes on both cards (`POST-FLR: verified (selectors survive AON)`); at handoff both cards report `cmd=0x0003`, `devsta=0x0000`, `lnksta=0x1103` (Gen3 x16), `pmcsr=0x0008` / D0 — **at the instant of handoff both cards are on the bus and in a clean state**; - The card that is unusable in Windows is `03:00.0` (Gigabyte, the non-console card): its `PciIo RomImage` is empty. The usable one is `04:00.0` (Asus, `SUBSYS_88181043`), whose `RomImage` is 148992 bytes a...
junkjunker • Nov 4, 2026
Please forgive my ignorance, I don't understand what to do with the v2.1.11-diag1 files. Thank you
physicsdolphin • Nov 4, 2026
**The build has been replaced in place — please download it again**: the file name is still `BOOTX64.EFI.2.1.11-diag1`, but a different binary was swapped into the pre-release (SHA-256 `13c9cc6e60cae182d9ce193eaaf5b116cb20851358f457c0406084afddb9bc7d`); the one some of you downloaded earlier had read-only sampling only and no canary. The canary adds the dimension of "was the card reset after handoff" — `POST-GOP: unlock intact` only says the write is still there; it cannot tell "nobody touched it" from "someone touched it and then it was reset". After the FLR verification the carrier writes two magic numbers into the PBUS software scratch and reads them straight back (no factory driver writes these two cells, and FLR/SBR/power removal all clear them); `POST-GOP`, `HANDOFF-OUT` and in-OS `ssdiag` each report once; interpretation must be paired with FEAT, table in §3. `ssdiag` must also be switched to **0.6.5** (attached to the same pre-release, run as administrator); the old version o...
physicsdolphin • Nov 4, 2026
@junkjunker no problem at all — here is the whole thing click by click. There is nothing to program and nothing you can break permanently; a normal reboot always brings the card back. **What we are doing and why:** we run the unlocker **twice, once per card**, so the two cards cannot interfere with each other during the resource handoff. Round 1 unlocks only the Gigabyte card (`03:00.0`, the one that disappears). Round 2 unlocks only the Asus card (`04:00.0`). Seeing only one card per run is intentional. Then we compare: if round 1 also makes the card disappear, it is that slot or that card; if both rounds are clean on their own, it is the two cards fighting over resources. **Time:** about 30 minutes, two reboots. **You need:** the same USB stick from before (already prepared), admin PowerShell, and a phone or the Screenshot tool. ### 1. Download the two files Open these links and let them save into `Downloads`: - Build: https://github.com/physicsdolphin/nv-uefi-unlocker/releases/...
physicsdolphin • Nov 4, 2026
I translated the reply into English and AI checked, hope that help you go through it. @junkjunker

Adjacent Repository Pain Points

Other highly discussed features and pain points extracted from physicsdolphin/nv-uefi-unlocker.

Top Replies
liuyansheng114 • Oct 30, 2026
补几条自我修正和新增实测，供下一个人少走弯路。原贴的读数结论不变，变的是对 的对照方式，以及一条我推得太远的结论。 ## 一、与 的对照取错了参照物 的日志开头是： ```text LOCAL-GA10X: scrubber skipped, ...
liuyansheng114 • Oct 30, 2026
补一条两分钟判据、一处收回，以及一处我自己也读错了的坑。 ## 一、ROM BAR 未打开 = 固件从未读过这张卡的 VBIOS 在 UEFI 环境下枚举一遍 PCI（工具只调 `LocateHandleBuffer` / `HandleProtocol` / `Pci.Read`...
liuyansheng114 • Oct 30, 2026
补最后一条实测：手动派发 option ROM 已经跑过，负结果；这条负结果反过来否掉了一整类方案。末尾有一个问题想问维护者。 ## 一、手动调度这张卡的 ROM：已执行，装不上 GOP 做法（EFI 应用，全程不写 flash）：...
Top Replies
physicsdolphin • Oct 31, 2026
这是小显存的第一个问题案例，请问你有尝试过正式版release 2.1.6 build也是不行的对吗？
buyuan-111 • Oct 31, 2026
> 1. 这是小显存的第一个问题案例，请问你有尝试过正式版release 2.1.6 build也是不行的对吗？ 是的，**正式版 v2.1.6 release build 也不行**，而且那是最早跑的一轮： | 轮次 | 构建 | marker | 结果 | |---|--...
buyuan-111 • Oct 31, 2026
> 1. 这是小显存的第一个问题案例，请问你有尝试过正式版release 2.1.6 build也是不行的对吗？ **补充 Windows 侧读数：事务成功，但状态没活过驱动初始化。** 同一次启动内，UEFI 阶段确实解锁成功（日志：`FEAT...
Top Replies
physicsdolphin • Oct 28, 2026
只有4GB显存可能是问题，之前没有试过任何一个
physicsdolphin • Oct 28, 2026
**先说结论：这不是显存 ≤4GB 的问题——从你这份日志看，payload 大概率已经执行成功、FEAT_PLM 门已经被砸开了。可以直接在 Windows 里验证并补写，不用重跑 U 盘。** ## 为什么这么判断 1. **门已经开了，而且扛...
hartienyu • Oct 28, 2026
按要求重新运行 USB，进入 Windows 后立即保持独显渲染并采样。 UEFI 的 HANDOFF-OUT 显示 FEAT_PLM=FFFFFFFF，但 Windows 有效读取为 FFFFFF8F，仍为 NOT_UNLOCKED。 已执行 --reapply，返回 reapply REFUSED: F...
Top Replies
physicsdolphin • Oct 29, 2026
你好，请阅读对应的release-note后尝试pre-release2.1.6，即可明白这是什么兼容性问题，这一版本对2080Ti 22G理论上修复了这个问题但对方还没有测试，你的问题和2080Ti 22G遇到的完全同构，请参见https://github....
physicsdolphin • Oct 29, 2026
最可疑的应该还是mailbox0=0x11这个，我仔细看看. 0x91应该和2080Ti 22G一样，是
physicsdolphin • Oct 29, 2026
> 2\. **`BL at IMEM dst=0xfe00` vs 参考机的 `0xfd00`**（代码里自标的异常）。 > 若 TU102 的 booter 对 BL 落位/BOOTVEC 有硬要求，这是目前最可疑的一条线。 > 本机 `hwcfg=0x7f7`、`chip=0x162000a1`、`FBI...
Top Replies
KakaruHayate • Nov 1, 2026
## 同型数据点：RTX 2070 `1f07` + 驱动 **630.20** / GSP **630.03**，Windows 侧同样被 **PF FLR** 复位 本机 v2.1.8 的 EFI 事务完整成功，进 Windows 后同样在 `nvlddmkm` 初始化阶段被复位，`ssdiag` 回到 s...
physicsdolphin • Nov 1, 2026
向日葵是个可能的高风险项，但我还没有做消融实验，我另一个有向日葵的电脑解锁完进windows 10就在内核早期蓝屏
physicsdolphin • Nov 1, 2026
https://github.com/physicsdolphin/nv-uefi-unlocker/issues/55 > 2\. 向日葵是个可能的高风险项，但我还没有做消融实验，我另一个有向日葵的电脑解锁完进windows 10就在内核早期蓝屏

Frequently Asked Questions

Market intelligence mapped to The `nv-uefi-unlocker` tool, specifically its multi-GPU detection and unlocking functionality within a UEFI environment, leading to incorrect GPU enumeration in the OS..

What problem does The `nv-uefi-unlocker` tool, specifically its multi-GPU detection and unlocking functionality within a UEFI environment, leading to incorrect GPU enumeration in the OS. solve?
Based on our AI analysis of the original developer request, its primary technical positioning is: Reliable, comprehensive NVIDIA GPU unlocking across multiple GPUs and diverse EFI-bootable systems, ensuring all detected hardware is correctly exposed and functional within the operating system.
Are engineers actively discussing The `nv-uefi-unlocker` tool, specifically its multi-GPU detection and unlocking functionality within a UEFI environment, leading to incorrect GPU enumeration in the OS.?
Yes, we have tracked 9 direct responses and active debates regarding this specific topic originating from GitHub Issue.
What architecture is tied to The `nv-uefi-unlocker` tool, specifically its multi-GPU detection and unlocking functionality within a UEFI environment, leading to incorrect GPU enumeration in the OS.?
Our proprietary extraction maps The `nv-uefi-unlocker` tool, specifically its multi-GPU detection and unlocking functionality within a UEFI environment, leading to incorrect GPU enumeration in the OS. to adjacent architectural concepts including nv-uefi-unlocker, UEFI pre-driver, NVIDIA GPU unlocker, Windows 10.

Engagement Signals

9
Replies
open
Issue Status

Cross-Market Term Frequency

Counts stored text matches for foundational terms like Windows 10 and ss1 in analysed public discussions. These counts do not measure adoption.

Stored text matches, not adoption or market-share measurements.

  • Windows 10: 9 records
  • ss1: 2 records
  • nv-uefi-unlocker: 1 records
  • UEFI pre-driver: 1 records
  • NVIDIA GPU unlocker: 1 records
  • dual RTX 3060 (GA106): 1 records
  • EFI-bootable system: 1 records
  • loader::pci: ENUMERATE: 1 records
  • NVIDIA display adapters: 1 records
  • dev=10de:2504: 1 records
  • tx-select-bdf.txt: 1 records
  • ACCESS_DENIED: 1 records
  • multi-GPU pass complete: 1 records
  • DMA released: 1 records
  • SELECTOR RESULT: 1 records
  • HANDOFF-OUT: 1 records
  • ss0: 1 records
  • readout1: 1 records
  • feat_plm: 1 records
  • wpr2_hi: 1 records
  • pmcsr: 1 records
  • dstate=D0: 1 records
  • tu10x: 1 records
  • ga/ada:ss0_unlock=true: 1 records
  • ss1_unlock=true: 1 records
  • plm_open=true: 1 records
  • loader::chainload: 1 records
  • PE header probe: 1 records
  • \EFI\Microsoft\Boot\bootmgfw.efi: 0 records