Speed deprecation? #1310
Replies: 4 comments 21 replies
-
Can you give us some details on what your settings are? We don't even know your OS, what memory/attention you have selected, etc. Kinda hard to even begin with "it's slower". |
Beta Was this translation helpful? Give feedback.
-
|
So what
Interesting... So what do you think Vlad, should we be switching to this dog text optimization method or is there a method to the madness with the sub-quadradic? Let's assume we have different video cards/setups. (Really appreciate the work Vlad I hope f1 was a damn blast) |
Beta Was this translation helpful? Give feedback.
-
|
Just FYI, Doggettx crashes the app when upscaling on my Mac and I may not be the only one (both in Automatic1111 and Vlad's UI). InvokeAI albeit tiny bit slower seems more stable. |
Beta Was this translation helpful? Give feedback.
-
|
i primarily test on nvidia and which settings are optimal for other platforms comes from community. |
Beta Was this translation helpful? Give feedback.
-
|
I might just edit the title to focus on memory attention. I haven’t gotten a chance to do more a-b testing since my last post. I am curious to see if changing to sub-q on Auto’s will give the same results, I don’t recall ever seeing my speeds tank that much but I don’t really stray away from what works. Hope to have more results later. Maybe a poll for amd users one which attention works best for them, no idea. |
Beta Was this translation helpful? Give feedback.
-
|
BTW I set the cross attention back to doggettx on my mac and it's not crashing during upscale anymore. One of the updates within the last week or so had to fix it. |
Beta Was this translation helpful? Give feedback.
-
|
same here on the last update - speeds go down at least 2times slower
|
Beta Was this translation helpful? Give feedback.
-
|
posts with "same here" are pretty much never a good idea. i don't even know what gpu you have. |
Beta Was this translation helpful? Give feedback.
-
|
my gpu - same here -i meant that yesterday my speed was 2times faster after update is slower on base pass and high res on same settings |
Beta Was this translation helpful? Give feedback.
-
|
which means its a completely different issue and most likely related to #1285 |
Beta Was this translation helpful? Give feedback.
-
|
actually i reinstall my drivers 2 days ago for this version |
Beta Was this translation helpful? Give feedback.
-
|
but yesterday for example on img 704x384 speed was around 6/it with same loras, without lora and embedings first pass around 8-9 it - today after update i get 2.5/it - live preview set on full vae |
Beta Was this translation helpful? Give feedback.
-
|
again, totally different issue than original poster. and there were no changes in core processing in the last few days at all. |
Beta Was this translation helpful? Give feedback.
-
|
One thing that I did today I just install this git https://github.com/bmaltais/kohya_ss - I am just an artist :) sorry Vlad maybe this git is not related, I'll play with settings and eventually reinstall the repository |
Beta Was this translation helpful? Give feedback.
-
|
Had a similar issue too. Live preview was set to nothing and it was ignoring what i was setting it to. Pushed a commit (eaeb545) for this and It's now using the preview I set instead of ignoring it. |
Beta Was this translation helpful? Give feedback.
-
Changing the live preview helped me too, I forgot I changed it to TAESD (don't even know what it is no idea why I changed it). That messed with my VAE it seems. 3070 8gb here |
Beta Was this translation helpful? Give feedback.
-
|
Curious to test that out |
Beta Was this translation helpful? Give feedback.
-
Keeping this short. I have a 6700xt and my average it/s at 512x512 using stock settings have always been 6.50 it/s. I've been using Auto's SD for months with the same speed. I had used Vlad's fork a month or so ago, and had also gotten roughly the same speed. After a month of using Auto's SD, I decided to give this fork a shot again especailly with all the new bells and whistles. However, my speeds at stock 512x512 is at 5 it/s! I reinstalled again from scratch, manually reinstalled torch and unchecked a bunch of settings such as live preview, but I am still only getting 5 it/s. Anyone who is up to date on this fork can shed some light on this? Is this a known issue? I don't want to open this as an issue unless others are having the same experience.
Beta Was this translation helpful? Give feedback.
All reactions