GPU Clock Throttle Idle is Active when desktop Locked (Windows+PyTorch)

I do PyTorch training with two independent processes on GPU 0 and GPU 1 on a Windows 11 machine with 2x3090. The scripts are limited to the corresponding GPUs using CUDA_VISIBLE_DEVICES env var. The monitor is attached to GPU 1. I am connected via RDP. When I disconnect from the machine, GPU 1 that has a monitor attached halves its performance in the training script.

In nvidia-smi -q I see “Active” next to the “Clocks Throttle Reasons” “Idle” subsection for GPU 1. It shows 100% GPU utilization, but seems limited to 150W.

In contrast the GPU 0 continues chugging at 350W, 99% utilization and “Clocks Throttle Reasons” “SW Power Cap”.

Attached the full output of nvidia-smi -q for GPU 1. WSL is not involved. Training script is Windows-native.
nvidia-smi.txt (6.6 KB)

Just for completeness - would it be possible to connect the monitor to GPU 0, and see if the throttling sticks with GPU1 or the problem follows to the GPU with the monitor… Hopefully that information will trigger some ideas with some of the folks who are more familiar with expected behavior .
Thanks and good luck.

Sorry, I am wrong. The GPU 1 no longer has monitor connected, but in the smi output “Display Active” is “Enabled” on it. It is “Disabled” on GPU 0.

I can not connect a monitor as the computer is in a remote location.