Find your next open-source project

Search by name, purpose, or keyword — or paste a GitHub repository URL

kate-land/ga100vfi

Hardware & IoT

Hardware, firmware and host tools for PWMVID voltage-glitch experiments on GA100 GPUs.

127
49
C
MIT
3 days old

What this is

This repository packages a complete PWMVID voltage-glitch experiment setup: a KiCad control-board design, STM32G431 firmware with a custom USB bootloader, host-side utilities, GA100 GPU0 register capture and analysis code, and a written experiment loop. It targets hardware researchers and engineers who want to build the board, run PWM waveform experiments, and then inspect how GPU register state responds after an attempt.

Each area lives in its own directory with dedicated documentation: hardware covers the P8 schematic, PCB, test-pad map and P9 assembly references; firmware covers the STM32 application, bootloader, host utilities and tests; driver/pwmvidctl holds the operator CLI, sequence planner, calibration profiles and waveform examples; gpu0-registers holds the register maps, bounded capture engine and offline analysis; and glitch-loop describes the experiment-loop pseudocode and historical commanded waveforms.

Offline checks

The README gives a single command block that exercises the firmware tests, bootloader tests, CLI unit tests and register-analysis workspace, run from the repository root:

(cd firmware && sh tests/run_tests.sh && sh bootloader/tests/run_all.sh)
(cd driver/pwmvidctl && python3 -m unittest discover -s tests -v)
(cd gpu0-registers && cargo test --workspace --locked)

These checks validate the software side only. Building the STM32 images requires PlatformIO, and the build environments are listed in the firmware documentation. Hardware bring-up is a separate step and depends on the measurements described in the bench acceptance document; no single command produces a working board.

Constraints

  • The register-analysis workspace builds and tests with Cargo using the locked dependency set, so a Rust toolchain is required for that component.
  • The CLI tests run under Python 3's unittest.
  • Firmware images need PlatformIO and the STM32 build environments from the firmware README.
  • A working setup ultimately depends on physical hardware: the control board, test pads and bench measurements defined in BENCH_ACCEPTANCE.md.
  • Original material is MIT-licensed; third-party terms and source notices are collected in LICENSING.md.

HotRepo's take

The project entered the new-project ranking on 2026-11-12 with 116 stars and 46 forks, recording 16 stars and 3 forks gained in the roughly 2.5-hour measurement window. The star and fork counts are unusually close together, which points to people cloning or building on the hardware design rather than just bookmarking it, though a single snapshot cannot establish a longer trend. It has no recorded position on the all-language daily GitHub rank. The scope is narrow and low-level — schematic, firmware, CLI and register analysis for one GPU generation — but the repository is structured enough to be reproduced by someone with the board and bench equipment.

Written by AI from the project's README, updated . The project's own docs are the source of truth.