CUDA OOM error and VRAM is not in use

[EQXai]

CUDA error: out of memory
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1
Compile with TORCH_USE_CUDA_DSA to enable device-side assertions.

I'm getting an OOM error with a 5090 with 32GB VRAM, the strange thing is that when running WF the VRAM never fills up, it never goes beyond 12GB and then gives an OOM error. This only happens with SkyReels, normal Hunyuan works perfectly.

[kijai]

Probably due to CFG, as the default behavior is to do cond and uncond batched in one pass which considerably increases VRAM use. I have added option to disable that to do it sequentially instead:

[EQXai]

Probably due to CFG, as the default behavior is to do cond and uncond batched in one pass which considerably increases VRAM use. I have added option to disable that to do it sequentially instead:

I have tried it and it doesn't work, could it be due to some incompatibility with the RTX 5000?

[kijai]

Hard to say, I do have 5090 on another setup and it's working fine so far.

[jwoeifjofwefawsfasd]

same OOM error with rtx 5090 @EQXai were you able to get any progress?

[EQXai]

same OOM error with rtx 5090 @EQXai were you able to get any progress?

No, but it could be anything, there are still a lot of things that are not supported by the 5000.