Possible memory leak?
I'm using SDXL Turbo with TAESD Decoder with RTX 3090 24gb vram and 64gb ram, windows 10.
At the start of SDXL Turbo, taking in a webcam, it would be .12s per execution. After about a minute or so, it would dip to 0.14, and eventually 0.24 or worse.
I'm not sure what's going on, possibly something is getting cached? Possibly on the "backend" python side?
Are you actually seeing memory usage increase? Does it happen with the regular VAE?
Can confirm this, there is a memory leak. Using wise memory optimizer to prevent browser from crashing. No errors when starting ComfyUI. System = 32 GB RAM and only a browser open with 5 tabs, Voicemeter app and wise memory optimizer app. Update ComfyUI every day. It returns to normal very slowly possibley every 5 to 10 seconds it drops 1%.
i could be wrong, everything seems to be working at the moment. Possibly just had to many nodes in the workflow. still strange because i bypass them when not in use.