Apple unified memory limits

[circuitwire58]

Thought I'd post a small recipe because a lot of threads ask similar things.

This isn't revolutionary, but the combination of Ollama and 4090 felt surprisingly usable once I tuned around continuous batching.

Feel free to tear it apart if you see something dumb.

[mlxgarden]

Did a small comparison this afternoon. On Threadripper box with MLX, I saw basically the same behavior.

[visionyard23]

I tested this tonight. The issue went away once I changed one setting and restarted the stack cleanly. The main thing that moved was latency, not quality.