Kacper Michajłow

> ROCm 6.4.1 introduces support for the RDNA4 architecture-based Radeon AI PRO R9700, Radeon RX 9070 XT, and Radeon RX 9060 XT GPUs for compute workloads. See https://rocm.docs.amd.com/projects/install-on-linux/en/docs-6.4.1/reference/system-requirements.html RDNA4 is...

> VAE decoding OOM issue is tracked here https://github.com/ROCm/ROCm/issues/4742 Actually OOM seems to be caused by forcing upcasting from bfloat16. ``` diff index 44aff376..5a0e4fd8 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@...

> Bug in windows and anyway, the new option is lossy garbage. Is there anything actionable on mpv side? Or we can go ahead and close this issue?

> good point whats the name of the lib that does the subsampling lav filter? do you mean swscale? Anyway, send input file used to produce this result.

> Both swscale API and internals are being reworked right now anyway, also the ability to even subsample in the first place is only offered for testing and/or to ensure...

We should have `--no-state` and make `--no-config` do what it actually should do, instead of bundling everything there. Either way, as it currently stands, `--no-config` should not save any history....

Does it happen with `--vo=gpu-next` also try `mpv-git`, there was small change in libplacebo, but I doubt it is affecting you.

We will need confirmation if white line artifacts are the same issue as RDNA4 corruption.

The patch is in [v6.16-rc3](https://github.com/torvalds/linux/releases/tag/v6.16-rc3) if someone wants to test or otherwise wait for stable release.

Can you report it upstream? There is nothing we can do about this on mpv side.