RTX 5090 on Lenovo Legion Pro 7i Gen 10 stuck at 95W; nvidia-powerd fails with “AutoFL: >1 CPU package” although kernel topology shows one CPU package

Graphics / Linux Linux
1

Hello,

I’m reporting what appears to be a Linux driver / nvidia-powerd issue affecting GPU power limits on a Lenovo Legion Pro 7i Gen 10.

System:

  • Laptop: Lenovo Legion Pro 7i Gen 10

  • CPU: Intel Core Ultra 9 275HX

  • GPU: NVIDIA GeForce RTX 5090 Laptop GPU (24 GB GDDR7, advertised max power limit 175 W)

  • OS: Ubuntu 22.04

  • Kernel: 6.8.0-107-generic

  • Driver: 580.126.09

  • CUDA reported by nvidia-smi: 13.0

Problem:
On Linux, the GPU appears capped at ~95 W under load. Under glmark2 and other GPU load, nvidia-smi reports power usage around 94–100 W and does not rise toward the 175 W max limit.

Relevant observations:

  • nvidia-smi -q -d POWER shows:

    • Current Power Limit: 95.00 W

    • Requested Power Limit: 95.00 W

    • Default Power Limit: 95.00 W

    • Max Power Limit: 175.00 W

  • nvidia-settings -q DynamicBoostSupport returns 1

  • nvidia-smi -pl 175 fails with “Changing power management limit is not supported for GPU”

  • AC power is connected (/sys/class/power_supply/ADP0/online returns 1)

  • BIOS graphics mode is set to Discrete Graphics

  • Lenovo performance mode has been enabled from Windows

nvidia-powerd:
I manually installed and enabled nvidia-powerd.service because the binary existed but the service unit was not installed by default. The service now starts and establishes a successful DBus connection, but then logs the following error:

ERROR! Failed to create AutoFL core: AutoFL: >1 CPU package

Current service status:

  • nvidia-powerd.service is active/running

  • DBus connection is established

  • The remaining error is the AutoFL one above

Important topology detail:
At first lscpu appeared to show 24 sockets, but deeper checks suggest the actual kernel-facing topology is sane:

  • /sys/devices/system/cpu/cpu*/topology/physical_package_id is 0 for all CPUs

  • /proc/cpuinfo shows physical id 0

  • numactl --hardware reports 1 NUMA node

  • dmidecode -t processor reports one processor package, 24 cores, 24 threads

So this does not appear to be a real multi-socket system. It looks like either:

  1. nvidia-powerd / AutoFL is misdetecting CPU topology on this platform, or

  2. some Linux platform support needed for Dynamic Boost on this machine is incomplete.

Expected behavior:
I would expect the GPU to be able to exceed the 95 W base limit and use Dynamic Boost toward the 175 W max power limit under load.

Actual behavior:
The GPU remains around 95 W, and nvidia-powerd logs:
ERROR! Failed to create AutoFL core: AutoFL: >1 CPU package

I have attached nvidia-bug-report.log.gz.

Please let me know if there are additional diagnostics you would like me to collect.

nvidia-bug-report.log.gz (703.1 KB)