2026.09.05 — GPU temperature source

New: GPU temperature as a per-fan source

A fifth source alongside Disk, CPU, IPMI and SAS — same linear ramp, same highest-speed-wins arbitration.

Two backends, chosen automatically per card:

Backend	Used for	Cost
drm hwmon node	AMD (amdgpu), nouveau, Intel	free — cannot wake the card
nvidia-smi	NVIDIA's proprietary driver, which exposes no hwmon node	~25 ms

NVIDIA cards need the Nvidia Driver plugin for nvidia-smi. AMD and Intel need nothing extra. As always, no GPU detected means the section simply isn't rendered.

Cards are identified by PCI address or GPU UUID, never by index. Indexes and hwmonN numbers renumber when hardware changes; a UUID does not.

Defaults are 50–85 °C. GPUs idle far hotter than disks and typically throttle around 85–90 °C, so a disk-like 40–70 °C range would hold the fan at full speed permanently — the same reasoning behind the SAS source's 55–80 °C.

The GPU is polled at most once every 30 seconds, even on a shorter fan interval. Querying nvidia-smi keeps the card out of its deepest power-saving state, so a 10 s fan must not drag the GPU along with it. This deliberately relaxes the usual "cache is always fresher than the fan's interval" rule: a fast fan may act on a GPU reading up to 30 s old. That is the intended trade, and it's noted in Help & Tips.

Verified on two machines — an RTX 4000 SFF Ada read by UUID through nvidia-smi, and a Radeon RX 6400 read by PCI address through its amdgpu hwmon node.

Install / update

Plugins → Check for Updates, or:

https://raw.githubusercontent.com/techanonymous/fancontrolplus2/main/unraid/fancontrolplus2.plg