DGX Spark (GB10) – Fans do not spin in headless boot mode, temperature rises to ~70°C

Hello NVIDIA DGX Team,

I am observing abnormal fan behavior on a DGX Spark (GB10) system.

System temperature gradually rises to ~60–70°C (no workload) and fans do NOT spin when booting in headless mode (no HDMI monitor connected).

If an HDMI monitor is connected and the system is rebooted. fans operate normally.


System Information

  • Model: DGX Spark (GB10)

  • DGX OS: 7.4.0

  • Driver: 580.126.09

  • CUDA: 13.0

  • Kernel: 6.17.0-1008-nvidia

  • BIOS: 5.36_0ACUM018

  • Architecture: aarch64

Output Log


Reproduction Steps

  1. Power off the system.

  2. Ensure no HDMI monitor is connected.

  3. Boot the system (headless).

  4. Leave system idle (no workload).


Observed Behavior

  • Fans do NOT spin.

  • No airflow detected at intake or exhaust.

  • System temperature gradually rises to ~60–70°C.

  • Chassis becomes noticeably hot.

  • nvidia-smi shows GPU in P8 (~3W idle).

Example:

GPU 0: P8, 3W, ~55C

Even when system temperature approaches ~70°C, fans remain inactive.


If HDMI Monitor Is Connected

If an HDMI monitor is connected and the system is rebooted:

  • Fans operate normally.

  • Airflow is present.

  • System temperature stabilizes around 35–40°C.

  • nvidia-smi shows normal behavior.

  • gnome-remote-desktop-daemon (user session) appears in process list.


Key Observation

The issue appears specifically tied to headless boot.

Fan control seems dependent on display initialization state.
In headless mode, fan PWM control does not activate even as chassis temperature rises significantly.

Expected behavior would be:

Fan activation based on thermal sensors, independent of HDMI/display initialization.


Questions

  1. Is DGX Spark fan control tied to display engine initialization?

  2. Is this expected behavior in headless mode?

  3. Is there a firmware/BMC thermal profile issue in headless boot?

  4. Is there a recommended configuration for proper headless operation?


Please let me know if additional logs are required.

This is not expected behavior and I cannot reproduce this. Can you check if your graphical manager is still active while in “headless” mode. It is expected to be active
systemctl status display-manager
You can also use nvtop to monitor what resources your unit may be using.

1 Like

FWIW – I run my dual setup in headless mode 24/7. It’s never connect to a monitor and I never see this behavior.

1 Like