version 0.34.0 and 0.34.1 throws an Not enough GPU memory" on AMD AI

When running any otherwise functional workflow in an instance running either version 0.34.0 or 0.34.1 I get the above error.
"Not enough GPU memory. Try reducing image resolution or batch size and run again"
My image resolution is 1024 X 1024 and I am not running batches.

While I stay on instances running 0.33.3 or 0.33.4 they all run fine.

I am running ComfyUI Desktop in Windows on Asus Flow Z13 with AMD 395 Max+ with 128GByte unified memory, hard allocated 96 for GPU.

[Not enough GPU memory. part of log.zip](https://github.com/user-attachments/files/31514540/Not.enough.GPU.memory.part.of.log.zip)

Start by comparing the supplied log excerpt with runs on versions 0.33.3/0.33.4 and 0.34.0/0.34.1 to identify where the GPU-memory error originates. Reproduce on the described Windows system with AMD 395 Max+ and inspect the relevant ComfyUI Desktop code and reports for AMD unified-memory handling. Done means the 0.34.x workflow runs without the false memory error; the issue does not name a file or test.

Written by the indexing model from the issue text.