You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
You won!
Well atleast Windows users have won, confirmed. Because I am currently on Windows. I think 560X might be the least powerful GPU currently with the weakest support. No one seem to talk about it, but I know a bunch of people have these gaming laptops with them and all of them had this budget entry level GPU (these competed with 1050Ti from Nvidia, and I know 1050Ti with its CUDA must be doing fairly better than AMD's trembling support lol).
Anyway, I was able to successfully run SD WebUI on Windows with DirectML. (Yes, it was the lshqqytiger's build)
For an 512x680 res image, with 20 iterations, I was able to finish renders in just under 1min when running --opt-sub-quad-attention --medvram --disable-nan-check --skip-torch-cuda-test --use-directml --update-all-extensions --upcast-sampling argument.
It jumps up to 1.5min or 2min when having --lowvram argument instead of --medvram.
Realistically, with medvram, you can only write like 60+60 tokens (positive prompts + negative prompts) or there about. And the renders are not always great.
With lowvram, you can go above 90+90 tokens easily but the renders take about 2 - 2.5 min.
Those arguments I mentioned above are the fastest settings. upcast-sampling doesnt seem to do much, but I just have it on there.
If anyone is interested in steps, maybe comment below, I might spend some time writing through the steps.
(PS - ROCm route didn't pan out for me with WSL2, and I didnt try ROCm windows native, partly because RX 560X is not on support list and there is no one talking about it. The couple I read seemed like a lot of time being dumped into it for questionable outcome and since I had other option, I didnt dive into that one.)
reacted with thumbs up emoji reacted with thumbs down emoji reacted with laugh emoji reacted with hooray emoji reacted with confused emoji reacted with heart emoji reacted with rocket emoji reacted with eyes emoji
-
You won!
Well atleast Windows users have won, confirmed. Because I am currently on Windows. I think 560X might be the least powerful GPU currently with the weakest support. No one seem to talk about it, but I know a bunch of people have these gaming laptops with them and all of them had this budget entry level GPU (these competed with 1050Ti from Nvidia, and I know 1050Ti with its CUDA must be doing fairly better than AMD's trembling support lol).
Anyway, I was able to successfully run SD WebUI on Windows with DirectML. (Yes, it was the lshqqytiger's build)
For an 512x680 res image, with 20 iterations, I was able to finish renders in just under 1min when running
--opt-sub-quad-attention --medvram --disable-nan-check --skip-torch-cuda-test --use-directml --update-all-extensions --upcast-sampling
argument.It jumps up to 1.5min or 2min when having
--lowvram
argument instead of--medvram
.Realistically, with
medvram
, you can only write like 60+60 tokens (positive prompts + negative prompts) or there about. And the renders are not always great.With
lowvram
, you can go above 90+90 tokens easily but the renders take about 2 - 2.5 min.Those arguments I mentioned above are the fastest settings.
upcast-sampling
doesnt seem to do much, but I just have it on there.If anyone is interested in steps, maybe comment below, I might spend some time writing through the steps.
(PS - ROCm route didn't pan out for me with WSL2, and I didnt try ROCm windows native, partly because RX 560X is not on support list and there is no one talking about it. The couple I read seemed like a lot of time being dumped into it for questionable outcome and since I had other option, I didnt dive into that one.)
Beta Was this translation helpful? Give feedback.
All reactions