Ollama now supports hardware acceleration on Apple M5 chips if 32 GB of RAM is installed

Ollama – an application for locally running AI models is now faster on Mac chips M5

What it looks like
New version: Ollama 0.19
Supported OS: Windows, macOS, Linux
Hardware acceleration: Only for Apple M5 chips (M5 Pro, M5 Max)
Memory requirements: Minimum 32 GB unified RAM

What changed
Ollama allows users to download and run AI models directly on their computer. Unlike cloud services such as ChatGPT, no constant internet connection is required.

Starting with version 0.19 the developers added support for Apple MLX – a machine‑learning framework – and for Unified Memory Architecture. This speeds up model performance on new Apple M5 chips, reducing time to first token and increasing overall generation speed.

How it works
* Neural accelerators: Ollama taps into specialized neural network blocks inside the M5 chips, making token inference faster.
* Models: Both personal AI agents (e.g., OpenClaw) and larger systems – OpenCode, Anthropic Claude Code, OpenAI Codex, etc. – run faster.

Limitations
1. Acceleration is available only on Apple M5 chips (M5 Pro/Max).
2. The computer must have at least 32 GB of unified memory; otherwise acceleration will not work.

Thus, if you have a Mac with an M5 chip and sufficient RAM, you can noticeably boost the performance of the local AI application Ollama.