V1 faster than V2

#8
by Shero-Safwat - opened

V1 is make 1 image per 55 seconds (ComfyUI) but for some reason V2 make it in 3 mins I don't know why

Shero-Safwat changed discussion title from V1 fater than V2 to V1 faster than V2

How much VRAM do you have? It's probably swapping between RAM and VRAM because V2 is slightly bigger.

RTX 3070 8 VRAM
but in comfyui the vram usage is 98% durian the rendering

RTX 3070 8 VRAM
but in comfyui the vram usage is 98% durian the rendering

That's the problem. Not enough VRAM means the model swaps constantly with RAM, slowing it down.

Thank god I'm not the only one with this issue!

RTX 3070 8 VRAM
but in comfyui the vram usage is 98% durian the rendering

That's the problem. Not enough VRAM means the model swaps constantly with RAM, slowing it down.

But wasn't this supposed to be lighter? In my case even full dev version produces images faster than v2 and I also have 8 VRAM

Thank god I'm not the only one with this issue!

RTX 3070 8 VRAM
but in comfyui the vram usage is 98% durian the rendering

That's the problem. Not enough VRAM means the model swaps constantly with RAM, slowing it down.

But wasn't this supposed to be lighter? In my case even full dev version produces images faster than v2 and I also have 8 VRAM

No, not lighter.
"V2 is 0.5 GB larger than the previous version, since the chunk 64 norm is now stored in full precision float32, making it much more precise than the previous version. Also, since V2 does not have second compression stage, it now has less computation overhead for on-the-fly decompression, making the inference a bit faster."

so we can't use v2 for good?
and maybe v3 can fix it or that is not possible?

"V2 is 0.5 GB larger than the previous version, since the chunk 64 norm is now stored in full precision float32, making it much more precise than the previous version. Also, since V2 does not have second compression stage, it now has less computation overhead for on-the-fly decompression, making the inference a bit faster."

Ohh I see, I think I misunderstood what this project was all about lmao.
Still, it puzzles me how for me v1 takes 1 minute to generate an image, full flow1_dev takes 2 minutes and V2 is taking up to 6 minutes. That's odd.

another question.. if I install Forge should have any different in the speed?

forge下 V1和V2差异不大,comfyui上,V2非常慢,所以应该节点问题。
There is not much difference between V1 and V2 under Forge, but on Comfyui, V2 is very slow, so there should be a node issue.

Sign up or log in to comment