Oddbean new post about login | logout @b5abe72d nope, I still have M1 mac regarding local AI image generation: better wait for M3, I am not shure M1 and M2 support it , whether Apple provides drivers for that like CUDA
@efa2e20e Have you tried any of them on your M1? Currently, I use ComfyUI running on Linux with a 1660ti and works pretty well but can't do video & bigger stuff, mainly because of only having 6 GB VRAM. MacBooks have much more RAM & VRAM and decent GPU power so I imagine it should be pretty good. Have you checked out Pinokio for installing AI? https://pinokio.computer/
@b5abe72d nope, i haven't tried i am not convinced many AI drivers will work on M1/M2 my guess only new macs in 2024 with M3 will support CUDA
@efa2e20e It works, you don't need CUDA to have it work on M1 Macs. It's slow with SDXL but has decent speed with SD 1.5, depending on what parameters you set. This guy took 10 minutes but his settings were pretty high. You can do a lower res image and then just upscale it to save time. https://youtu.be/6LjDykRMtg8?si=XtfmQGcZffUB8wgp
@b5abe72d i am not convinced as long as no M3 macs my recommendation: just wait couple months for M3 macs
@efa2e20e You most likely will never have CUDA on newer Macs so long as Apple continues using their own chips. CUDA is Nvidia only & only works on Windows & Linux. Also doesn't help that CUDA is closed source & only developed & released by Nvidia. You will have to rely on Apple to release whatever their equivalent of CUDA, whenever & whatever that may be. You might not even need it, most of the libraries you need are open source, like PyTorch, Tensorflow, and more https://www.unite.ai/10-best-python-libraries-for-machine-learning-ai/
@b5abe72d PyTorch needs CauDA to run fast on GPU