Open wardensc2 opened 1 month ago
I try to replace all the checkpoint except VAE to GGUF to reduce Vram however my PC still crash and need hard reset when enter vae decoding process. Not sure what happen, I though the workflow frying my 3090 (temperature going up to more than 80 degrees, normally only around 70 when render image).
The VAE decoding is extremely heavy with these models, you should try enabling the vae tiling.
I try to replace all the checkpoint except VAE to GGUF to reduce Vram however my PC still crash and need hard reset when enter vae decoding process. Not sure what happen, I though the workflow frying my 3090 (temperature going up to more than 80 degrees, normally only around 70 when render image).
The VAE decoding is extremely heavy with these models, you should try enabling the vae tiling.
I try turn both vae tiling and vae slicing on but after turn it on. Crash happens more frequent.
Perhaps you can try to adjust the vae decoder node's settings to reduce VRAM requirement? With everything on default in this workflow, I was getting OOM on a 8Gb GPU, but these ones for CogVideo Decode node allowed me to obtain result: enable_vae_tiling true tile_sample_min_height 120 tile_sample_min_width 184 enable_vae_slicing true
I try to replace all the checkpoint except VAE to GGUF to reduce Vram however my PC still crash and need hard reset when enter vae decoding process. Not sure what happen, I though the workflow frying my 3090 (temperature going up to more than 80 degrees, normally only around 70 when render image).