Running out of VRAM on easy forLoopEnd - rest of flow dont use more than 60% VRAM

[supereth]

@kijai

Please advice - I've followed your advice of relying more heavily on the blockswap node on your loop workflow increasing the values and even activating the img embed, txt embed and use non blocking options, to the point that my VRAM usage ranges between the high 50-60%s throughout all the flow but at the time of the forLoopEnd step i am still getting the 'you ran out of memory on your GPU' wall.

I've got a 9800X3D on a ASRock X870E TAICHI with a T705 M.2 Gen5 and 64GB of RAM @ 6000MHz with a RTX 5080 16GB VRAM, how can I prevent this problem?

Thanks in advance.

[supereth]

Findings - After busting Gemini balls for quite a bit and many trial and errors I came to the conclusion that reducing the weight of the data incoming to the forLoopEnd process by using an Upscale_Latent_By node set to 0.5 with the intention of doubling it afterwards with another Upscale_Latent_By node set to 2 (with the subsequent VAE encoding and decoding and possibly loss of quality on the end result) after the cursed forLoopEnd could be a workaround to avoid the spike in the forLoopEnd process but I have still not been able to succeed.

Deploying some Clean_VRAM_Used nodes in different places also didnt lead to fruition.

Calling brighter minds for help.

[GeorgeS2019]

Which workflow and what image size, so far longCat fails with 11gB vraM