Hi maintainers,

I'd like to gauge interest before opening a PR for a small Stats-panel addition.

What I'd add: three optional rows in the existing Stats dialog (frontend/widgets/OBSBasicStats.cpp) — GPU usage %, VRAM used/total, GPU temperature. Same color thresholds and update cadence as the existing CPU/Memory rows.

How it works: dynamic-loads nvml.dll (Windows) / libnvidia-ml.so.1 (Linux) at runtime via os_dlopen. No new build-system dependencies, no link-time changes. If NVML isn't present (non-NVIDIA host, missing driver, macOS), the rows display "—" and the code path is otherwise a no-op.

Scope I'm proposing for this PR (NVIDIA-only): I'd ship the NVML implementation now and design the surface so AMD ADL / Intel Level Zero backends can slot in later behind the same accessor. I don't want to scope-creep this into a vendor-abstraction PR on the first pass, but happy to discuss the shape if you'd prefer that up front.

My background here: I maintain a downstream fork that's been running this code since 2026-05-14. I'd handle the i18n through en-US.ini properly, run clang-format, follow CODESTYLE.md, and split the commits per the 50/72 / frontend: prefix conventions in CONTRIBUTING.md.

Disclosure: I draft code with AI assistance (Claude). I've been reviewing, building, testing, and iterating on this in my fork — including a code-review pass that fixed real bugs — and I'd own ongoing maintenance. Happy to discuss in more depth if the AI provenance is a blocker; not trying to slip anything past anyone.

Questions for maintainers:

1. Would you accept this feature at all, given Stats is mostly CPU/memory-focused today?
2. NVIDIA-only first pass acceptable, or do you want a vendor abstraction up front?
3. Anything about the dlopen approach you'd want done differently?

Screenshots of the current implementation in my fork will follow in a comment below.

Thanks for the time.