Accelerating Stable Diffusion Inference on Intel CPUs with Hugging Face (part 1) 🚀 🚀 🚀

[@GeekDynamicsLab]

Amazing work, Im interested in this awesome code, thanks for sharing and thank you to all hugginface contributors and amazing people that creates AI tools and models

[@juliensimonfr]

Our pleasure!

[@redkuri9333]

can you make some video using this method step by step to Stable Diffusion Web UI?

[@kenttom]

can you make a video on how to install SD on intel core i5 + intel iRIS xe graphics ? i heard that it is not possible . stable diffusion does not support intel

[@adrianmucci4860]

Thanks for sharing this information. What version of Xeon use for this? Intel xeon 2680v4 is good for this?

[@juliensimonfr]

You need a Sapphire Rapids CPU. Not sure about CPU names :)

[@carlosrfs99]

In a Xeon E5 2650 v4 with just one channel of ram been used, i had 35% of performance improvement.

[@DataKnight]

Thanks for sharing

[@juliensimonfr]

Sure!

[@synthoelectro]

Stable diffusion reddit would love to hear about this.

[@juliensimonfr]

Feel free to post it there, I'm not on Reddit.

[@synthoelectro]

@@juliensimonfr let them find it ;) Most don't use CPU but it is pretty neat you can do this at that speed.

[@Lifelessons-sv7pr]

Will it work on stabilityai/sd-turbo? I am unable to make it work 😢

[@juliensimonfr]

I don't know. You should ask for help at discuss.huggingface.co or create an issue in the Optimum Intel repository.

[@ericyang7694]

Is it possible to do this for windows?

[@pika9985]

is that work with 3rd gen or 4th gen cpu's ?

[@juliensimonfr]

Previous generations don't have AMX, so don't expect the same acceleration.

[@skimaskcitizen301]

Does this work with windows 10? or does it work only for mac?

[@juliensimonfr]

I haven't used Windows since 2007, so I honestly don't know :) I suppose it should work if you have a working Python environment?

[@bomar920]

How about apple m2 chips ?

[@juliensimonfr]

huggingface.co/docs/diffusers/optimization/mps

[@bomar920]

@@juliensimonfr of course, it would be nice to have small tutorial on this as well