Panchovix’s Vix tools load an edited NVIDIA VBIOS image without writing the GPU’s flash chip. Windows and Linux use different loading paths, and each has its own system requirements. The Windows loader and Linux loader describe how those overrides work. Reports from Tom’s Hardware and El Chapuzas Informático about the project are dated October 19, 2026.
The distinction matters: the modified image can change the settings used by the GPU, while the firmware stored on its flash chip remains untouched. That avoids writing an edited image to the chip, but it does not make aggressive power, current or voltage settings safe.
What the Vix tools change
The Vix Editor works with a VBIOS image offline and saves an edited copy. Separate loaders then apply that image through Windows or Linux. The image used by the driver or GPU System Processor (GSP) changes; the card’s stored firmware does not. The editor documentation describes settings that vary by architecture, including power, voltage, clock offsets, current and fan parameters.
The two loaders use different routes:
| Operating system | How the edited image is loaded | How to return to the normal path |
| Windows | The loader stores the image in the GPU driver’s registry key and restarts the GPU. | Use the loader’s unload function to remove the override. |
| Linux | The loader patches NVIDIA’s open kernel modules so GSP starts with the image in memory. | Choose the separate normal boot entry, which remains available. |
On Windows, the loader’s Verify function checks whether the driver is running the stored override. A clean driver installation or update can drop it, so the project advises checking the applied state after driver changes. The project’s Windows release executables are not code-signed and may prompt a SmartScreen warning.
Windows requirements and documented tests
The Windows loader requires 64-bit Windows 10 or 11, administrator access, an NVIDIA 616.x or 617.x driver at version 616.56 or newer, and an edited payload matching the GPU’s exact PCI device ID. The loader’s requirements and behavior make that device match a concrete prerequisite, not an optional convenience.
The project lists real-hardware examples for the GeForce RTX 5070, GeForce RTX 4090 and RTX A2000. Its Windows documentation marks other Ampere and Ada GPUs as untested. The editor also describes Turing editing and loading as experimental, so architecture coverage alone does not tell you whether a particular card and operating-system path has been tested.
Linux support is tied to a specific setup
The Linux loader uses NVIDIA’s open kernel modules and GSP to start with the edited image in memory. It creates a separate boot entry and retains a normal entry as a fallback. Its setup requirements include NVIDIA 615.x or newer with open kernel modules, Python 3.11 or newer and kernel headers; Secure Boot must be disabled or configured with a user signing key.
The project lists Fedora 43 with GRUB/BLS, dracut, DNF5 and NVIDIA 615.x open modules installed with the .run installer as supported on real hardware. Its Linux documentation also lists tests on the GeForce RTX 5090, 5080, 5070 Ti, 5070 and 5060 Ti, RTX PRO 6000 Blackwell variants, and RTX 2000 Ada. Several other distribution and installation combinations are marked experimental or unsupported.
Architecture and use-case limits
The editor’s settings vary by GPU architecture. For Ada and Ampere, GSP reads at most 512 KiB from the legacy image, so the payload needs relocation; the project labels relevant relocation paths experimental. The editor says VRAM timing changes do not take effect because the card reads those timings from physical firmware.
Laptop behavior has its own constraints. Dynamic Boost may still control effective GPU power, and laptop fan control belongs to the laptop rather than the VBIOS image. A higher configured power ceiling therefore does not necessarily change the power the GPU uses.
A no-flash override can still damage hardware
Panchovix’s project warns that power, current or voltage settings beyond a card’s design limits can damage hardware. It also says an edited image’s cryptographic signature no longer matches and warns against flashing that image with nvflash.
El Chapuzas Informático reported developer-configured power-limit examples of 600 W to 800 W for the GeForce RTX 5090, 360 W to 500 W for the RTX 5080, 250 W to 350 W for the RTX 5070, 185 W to 300 W for the RTX 5060 Ti, and 550 W to 671 W for the RTX 4090. Those are configured limits, not measurements of sustained power consumption or evidence of FPS gains, as the report notes.
