@ivanfioravanti: Apple M5 Max + MLX = raw power! Look at this demo I'm playing with "FasterLivePortrait-MLX" I started with MPS but resu…
Summary
The author demonstrates that migrating a LivePortrait implementation from MPS to Apple's MLX framework on an M5 Max chip results in significantly better performance and speed.
Similar Articles
@awnihannun: Pretty remarkable what two M5 Ultras + MLX can do now
Awni Hannun 展示了两台 M5 Ultra 配合 MLX 所能达到的性能水平，凸显本地 Apple Silicon 推理能力的显著提升。
@ivanfioravanti: Interesting video of M5 Max, on impact of Low, Automatic and High power modes on inference. - No external monitor attac…
A performance test demonstrates the impact of Low, Automatic, and High power modes on LLM inference speed on an M5 Max MacBook, showing significant differences in token generation rates and power consumption.
Nex-N2.5-mini-MLX-4bit on Apple M5 Max — 133.6 tok/s — llm-bench.io
Benchmark results for the Nex-N2.5-mini-MLX-4bit model on Apple M5 Max hardware, achieving 133.6 tokens per second generation speed and quality scores up to 85.80 in research tasks.
@jun_song: The new engine for MLX is in its final stages of development. Just ran GLM-5.2 on a single MacBook (116GB) hitting 41.8…
Jun Song announces the final development stage of a new MLX engine, achieving 41.8 tok/s on a MacBook with a 256k context window and only ~4% quality loss, representing a significant performance improvement.
Apple M5 isn't making full use of its matmul cores yet
Apple M5 silicon supports INT8 activations for matrix multiplication, but inference backends like MLX and Llama.cpp currently use 16-bit; custom w8a8 kernels achieve up to 1.4x speedup on Gemma4 prefill tasks.