I'm not sure how many people are using this card, but in case anyone was curious if it's capable of running these models and what the generation times look like for a baseline, I thought I'd post these.
System:
Windows 10 22H2
ComfyUI portable 0.42.1
RTX 3060 Ti 8GB VRAM
i7-11700 32 GB RAM
Used INT8 convrot quants for all models called out in the default templates from ComfyUI. Pre-loaded model on each run so model loading time from disk wasn't a factor.
Total time for each image:
Model 1.0 MP 1.5 MP 2.0 MP Total
Klein3 16s 24s 35s 75s
Krea2 24s 39s 53s 116s
Moss 53s 81s 122s 256sSeconds per megapixel (s/MP):
Model 1.0 MP 1.5 MP 2.0 MP Average
Klein3 16.0 16.0 17.5 16.5
Krea2 24.0 26.0 26.5 25.5
Moss 53.0 54.0 61.0 56.3Prompt:
a cute cartoon albino possum, wearing a black witch's hat, riding on a glowing jack-o-lantern. it has its mouth wide open in a yawn.
dark fantasy forest setting. crisp color palette. dynamic environment. film grain. depth of field.
Seed: 1000This isn't meant to be a quality comparison, the prompt was short and basic, but in case you're curious about the results, going from 1.0 to 1.5 to 2.0 for each model increasing the latent size, not using any upscalers, using the same prompt and seed.
Klein3
Krea2
Moss
