Tips for running smoothly on Apple Silicon?

[connorjewiss]

Followed the GPU Acceleration tips for running smoothly of Apple Silicon, but still laggy compared to Windows versions I've seen. Running on 32GB M1 Max.

[sporkman]

Same here, I'm at around 2 fps, what should I be seeing?

[briascoi]

Same here, and no one is answering this question. There is no help at all with Mac :(

[hacksider]

Sorry about that, I don't really have a mac to begin with so I can't answer it. And having a dedicated GPU is really the best way to run this with CUDA ☺️

[circlesmiler]

Ok... I make a wild guess: You need a lot of GPU power to run this smoothly. Apple Silicon Chips has integrated GPU which is OK for doing your work and playing some games. But there is a reason why gaming is still done a lot on PC + crazy graphic cards that have a power consumption like a hair dryer. :-) (PS: I'm a happy Mac user, but there are limitations)

[hacksider]

This, plus I don't have a mac to actually test it 😅

[cdrake118]

I think it is a bug because it utilizes 0% of GPU currently.

[hacksider]

I don't really have a mac to begin with to resolve the issue so we're at the mercy of the community help here. Plus there's really an advantage of having dedicated GPU.

[fakeneuron]

When looking at Activity Monitor, none of the work is being done by the GPU which runs at 0%. All is being done with CPU.

[manINsano-GIT]

same issue. looking for community support.

[dodgyrabbit]

The biggest improvement on mac (tested on M2 pro) comes from this:

Delete models/inswapper_128_fp16.onnx

Download this model instead:
https://huggingface.co/hacksider/deep-live-cam/blob/main/inswapper_128.onnx

Copy it to the models directory and rename it to inswapper_128_fp16.onnx

Then make sure you start with coreml execution provider, like so:

python run.py --execution-provider coreml

Should get around 4 fps

Note that this does not help you with the face_enhancer. That is a completely separate python module with it's own model and is not optimized for mac GPU. I believe it's all on CPU so you're going to get about 1 frame every few seconds if you turn it on.

[UltraSeeder]

Works good

[tomkirk]

this does improve it a bit thanks.

[tomkirk]

I also have this issue on an M3 pro.

• 11-core CPU
• 14-core GPU

I am also experiencing about 2fps.

Any further ideas on how to improve this?

[fakeneuron]

I have the same issue with M3 Macbook Air. When looking at Activity Monitor, i see all the work is being done by the CPU. GPU is at 0%. Looks like something in the code is not properly directing it to run on the GPU.

[awu0403]

Is this problem still unresolved at present? When looking at Activity Monitor, none of the work is being done by the GPU which runs at 0%. All is being done with CPU