Can you make one for my GPU?

#3
by MasterCeddy - opened

Hi friend, could you make one version compatible with this GPU?

ZOTAC RTX 4080 Trinity OC 16 GB
AD103, revision A1
Device ID: 10DE:2704
Subsystem: 19DA:1688
VBIOS: 95.03.0E.40.9A
SHA-256: 58088039EFB6EA35755624EB6292C216D37D00F551CACB3F8D01D625800A39DB

I followed your guide. It was hard to follow, and in the end I couldn't succeed despite trying several times with the help of AI

My GPU is rtx4080, non-super but only 16gb. Perhaps this is the problem

If you have this profile, please share. Thank you nevertheless

Specific conclusion for your card

Your checks established:

Plain Text

GPU-Z/NVGI container size: 2,048,000 bytes

Embedded PCI ROM offset: 37,888

Embedded ROM size: 879,616 bytes

Embedded signature: 55aa

Your embedded ROM SHA-256: d9f241148f7d457201a049067859c93e40f22e6cf2fc557b4fd68e8d9e8133c8

Checker result: NOT MATCHED

This means:

    Your GPU-Z export is not obviously malformed.

    The checker correctly found a valid embedded PCI ROM.

    The extracted ROM is structurally plausible.

    Its content does not match the released profile.

    Rebuilding the EFI does not adapt it to your ROM.

    Under the project’s own documented scope, you should not boot the resulting EFI.

The installation guide explicitly excludes different ROMs from the validated scope, while the README says the release is pinned to one device/subsystem and one specific ROM. [huggingface.co], [huggingface.co]
Recommended corrected workflow

The guide should be reorganized into this order:

Plain Text

    Confirm exact supported hardware and modification.

    Record PCI device ID, subsystem ID, VBIOS version and memory size.

    Export the ROM with GPU-Z.

    Run check_rom.py.

    If NOT MATCHED: stop. Do not build or boot.

    Install declared dependencies: Python 3.14, zstd and optionally Zig 0.14.1/GCC.

    Run prepare_assets.py.

    Verify every generated asset and hash.

    Use the included EFI, or optionally rebuild with a real Zig executable path.

    Verify the final USB tree and hashes.

    Record Secure Boot and recovery settings.

    Boot manually and follow a defined status/recovery decision tree.

    Benchmark with controlled repeated measurements.

Bottom line: the most consequential missing instruction is an unmistakable hard stop after NOT MATCHED. Everything after that point currently looks procedural, which can encourage users to keep going even though the profile is explicitly unvalidated for their ROM.

Hi @enn36 @MasterCeddy

Checker result: NOT MATCHED happens because this repository is hardcoded for a single modified 32 GB board profile and a pinned VBIOS ROM SHA-256. Standard 16 GB 10DE:2704 cards will always fail that static hash check.

I built an automated cross-platform implementation on top of TOTORONG's and Zagitta's upstream work that handles standard 16 GB RTX 4080 (10DE:2704, AD103) boards out of the box:
https://github.com/MindAllik/GPU-Tensor-Unlock-Universal

Instead of matching a hardcoded ROM hash or editing local_profile.h manually, Windows_Launch_UI.bat (core_detector.py) reads your card's PCI IDs (10DE:2704 / 19DA:1688), 16 GB VRAM window, and Windows ESP GUID directly from the system, compiles a board-matched BOOTX64.EFI, and runs a live pre-flight PromCheck (BAR0 + 0x300000 + dry-run FRTS patch in RAM) before disconnecting the display driver.

Desktop 10DE:2704 is currently marked as experimental_simulated in our matrix (my physical test bench is an RTX 4070 Ti 10de:2782). If you run it, post your EFI\UNLOCK\REPORT.TXT and the built-in benchmark numbers here or in GitHub Issues - both a clean pass and a pre-flight abort log are useful data for moving 10DE:2704 to hardware-tested, thank you!

Sign up or log in to comment