Hi everyone,
I have three GIGABYTE AI TOP ATOM systems, each running separately. On all three units I see the same behavior:
When I load an LLM with vLLM, the fans start spinning after a short time and then continue running indefinitely, even when there are no active vLLM requests and the machine is basically idle.
The temperatures look very comfortable. According to nvidia-smi, the GPU is at around 39°C and 9 W while idle.
Sensor readings:
mt7925_phy0-pci-90100
Adapter: PCI adapter
temp1: +40.0°C
mlx5-pci-20100
Adapter: PCI adapter
asic: +45.0°C (crit = +105.0°C, highest = +59.0°C)
mlx5-pci-0100
Adapter: PCI adapter
asic: +45.0°C (crit = +105.0°C, highest = +59.0°C)
acpitz-acpi-0
Adapter: ACPI interface
temp1: +41.5°C
temp2: +40.2°C
temp3: +40.1°C
temp4: +39.2°C
temp5: +39.9°C
temp6: +41.5°C
temp7: +39.8°C
mlx5-pci-20101
Adapter: PCI adapter
asic: +45.0°C (crit = +105.0°C, highest = +59.0°C)
mlx5-pci-0101
Adapter: PCI adapter
asic: +45.0°C (crit = +105.0°C, highest = +59.0°C)
nvme-pci-40100
Adapter: PCI adapter
Composite: +37.9°C (low = -5.2°C, high = +82.8°C) (crit = +84.8°C)
Sensor 1: +37.9°C (low = -273.1°C, high = +65261.8°C)
System utilization:
top - 12:58:04 up 1:57, 1 user, load average: 0.40, 0.23, 0.19
Tasks: 462 total, 1 running, 461 sleeping, 0 stopped, 0 zombie
%Cpu(s): 0.1 us, 0.2 sy, 0.0 ni, 99.6 id, 0.1 wa, 0.0 hi, 0.0 si, 0.0 st
MiB Mem : 124546.2 total, 5999.0 free, 115396.2 used, 4347.1 buff/cache
MiB Swap: 16384.0 total, 16384.0 free, 0.0 used. 9150.1 avail Mem
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
37444 root 20 0 147.6g 3.8g 729264 S 3.3 3.1 7:23.16 VLLM::EngineCore
36896 root 20 0 3164728 112876 41636 S 1.3 0.1 1:12.32 /usr/bin/dcgm-exporter
...
All three GIGABYTE AI TOP ATOM systems show exactly the same fan behavior. I did not experience this with my ASUS GX10 under a similar setup, so I suspect this might be firmware-specific fan-control behavior?
With the fans running continuously, each GIGABYTE AI TOP ATOM uses around 50 W at idle with the LLM loaded.
Does anyone know whether this behavior is expected? Is there a firmware or, driver setting to make the fans spin down again when the system is idle and temperatures are low?