Changelog
New Features
- 2f27d6c feat: add a multi-GPU overview dashboard
- f43b5b0 feat: add an experimental NVML collection backend
- 6b078b2 feat: add value ranges and a config file to the fake nvidia-smi
- cbef58d feat: expose GPU recovery action and fabric state as metrics
- 6bea5d0 feat: fluctuate values and simulate multiple GPUs in the fake nvidia-smi
- a354542 feat: harden the http serving and scrape handling
- c251821 feat: let the fake nvidia-smi override field values via --set
- 5482e48 feat: overhaul the Grafana dashboard
- 5640f54 feat: warn when the driver stops providing an NVML function
Bug Fixes
- 594e4f0 fix: record the symbol extraction tool and make the fatal error nil-safe
Documentation
- d6f659c docs: describe the use cases in the readme
- 911c455 docs: update the Grafana dashboard screenshots
Refactors & Internal Changes
- fcf2112 refactor: rework the all-in-one Windows install script
Dependency Updates
- 670f569 chore(deps): update dependency golang/go to v1.26.5
- 72dcd8f chore(deps): update dependency markdownlint-cli to v0.49.1
- f12304c chore(deps): update docker images
- a13e13a chore(deps): update github actions
- 7fa2645 chore(deps): update github actions
- 2918e0c chore(deps): update go packages
- f938c53 chore(deps): update module github.com/prometheus/common to v0.70.0
Build, CI & Chores
- 2feac44 chore: extend the compose dev stack for dashboard development
- 19bdbdc chore: lint the Grafana dashboard with the grafana dashboard-linter
- b51a2ca chore: resume winget publishing
- da0601d chore: satisfy the linters for the cgo build
- f9341ec chore: simulate two nodes in the compose dev stack
- f6977bb test: check driver library symbols against the recorded inventories
- b7e15a2 test: compare the two backends field by field on a real GPU
- 5eef198 test: detect driver drift against the recorded captures
- 47e983a test: tighten the drift checks after review
Other Changes
- de6404e capture: NVIDIA H100 80GB HBM3, plain and MIG (590.48.01, linux-x86_64)