Skip to content

v2.4.0

Choose a tag to compare

View all tags
@kinncj kinncj released this 01 Jul 19:40
v2.4.0
afa515f
This commit was signed with the committer’s verified signature.
kinncj Kinn Coelho Juliao
GPG key ID: 90479486A80D1E81
Verified
Learn about vigilant mode.

Accelerator coverage: multi-GPU, Intel NPU utilisation, and AMD fan speed.

Added

  • Multi-GPU NVIDIA hosts aggregate across cards. nvidia-smi emits one row per GPU; the collector reads them all — power.gpu sums (so power.total is right), gpu.util/gpu.clock/gpu.mem.util average, gpu.temp/gpu.fan take the hottest, and gpu.vram pools used/total with a (N GPUs) note. Single-GPU hosts unchanged.
  • Intel NPU utilisation. On Intel NPUs (intel_vpu — Core Ultra / Arrow Lake), npu.util is a real reading derived from the driver's cumulative npu_busy_time_us counter (idle = 0%). AMD XDNA and Apple ANE expose no counter and stay Unavailable-with-reason.
  • AMD gpu.fan from amdgpu sysfs. Reads the pwm1 duty cycle (0–255) from the amdgpu hwmon and reports it as a fan-speed percent, so a discrete Radeon shows fan speed without amd-smi. Absent on APUs with no dedicated GPU fan.
  • power.npu explains its dash — Unavailable-with-reason (no NPU power counter) where no source exists, mirroring npu.util.
Assets 43
Loading