hunyuan t2v VAE Decode Tiled stuck for a long time but Wan2.1 t2v VAE Decode Tiled is OK

The hardware and software environment are exactly the same.

CPU: i5-7500
GPU: AMD 7900XT 20GB
RAM: 32GB
Ubuntu24.04
Pytorch version: 2.7.0.dev20250308+rocm6.3
ComfyUI version:0.3.24

Video specifications:
480x960
20 steps
49 frames
teacache acceleration

hunyuan_video_t2v_720p_bf16.safetensors(25.6GB) will be stuck for a very long time and cannot be completed under VAE decode tiled (256 64 64 8).
VAE decode tiled can only work at the highest parameter of 128 64 32 8, and the node consumption time is about 55 seconds.

I just tested the Wan2.1 model: wan2.1_t2v_14B_bf16.safetensors(28.6GB)
It can successfully complete the video generation under VAE decode tiled (256 64 64 8), and the VAE decode tiled (256 64 64 8) node consumption time is only 22 seconds.

I'm sorry I don't know what information to upload, but I can help with the test.

Thank you

In addition, I replaced Pytorch2.6+ROCm6.2.4, and the same situation occurred.