-
Notifications
You must be signed in to change notification settings - Fork 4
Dynamic VRAM Issues #18
Copy link
Copy link
Open
Description
I'm currently convinced that your diffusion model loader node is breaking Dynamic VRAM for the checkpoint and therefore negating speed gains I would have gotten from that and resulting in generations taking longer than if I was using FP8.
I am using your int8 version of LTX 2.3.
Looking at my command prompt window, it seems to "work" but then there's a giant wall of pin errors and it's clear that it fails to use Dynamic VRAM and my generations are slower than FP8 on my RTX 3090.
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels