GMem reacts poorly but Compute Units maxxes out on AMD R4750G APU

### Expected Behavior

Image generation outputs to be out as usual as instructed by my workflows

### Actual Behavior

AMDGPU Compute Units maxxes out but GMem/UMA remains low activity

### Steps to Reproduce

(1) Test ComfyUI on AMD R4750G APU
(2) Max out UMA to 16GB in UEFI BIOS
(3) Set HSA_GFX Gen to 9.x; Run ComfyUI in `--highvram --disable-smart-memory`
(4) Try an arbitrary workflow and look at Plasma system monitor
(5) Take a screenshot and look at use of graphical memories and Compute Units
(6) If one of Compute Units or GMems are maxxed out but the counterpart reacts barely?

### Debug Logs

```powershell
got prompt
Requested to load SD3
Loading 1 new model
0%| | 0/20 [00:00 b i j', q[:, i:end].float(), k.float()) * scale
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/opt/local/ComfyUI_SDiffXL/pyenv/lib/python3.12/site-packages/torch/functional.py", line 385, in einsum
return _VF.einsum(equation, operands) # type: ignore[attr-defined]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.OutOfMemoryError: HIP out of memory. Tried to allocate 11.07 GiB. GPU 0 has a total capacity of 23.46 GiB of which 13.28 GiB is free. Of the allocated memory 9.76 GiB is allocated by PyTorch, and 99.83 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting PYTORCH_HIP_ALLOC_CONF=expandable_segments:True to avoid fragmentation. See documentation for Memory Management (https://pytorch.org/docs/stable/notes/cuda.html#environment-variables)

Prompt executed in 2.83 seconds
```

### Other

ASRock AMD X300 DeskMini on AMD R4750G APU × 16GB UMA
OS: Artix Linux using OpenCL-AMD 6.1.3
With HSA GFX Gen 9.x