mlx-vlm on the M3 Ultra 512GB
llamaperf has 4 mlx-vlm reports on the M3 Ultra 512GB, none of them a plain single-card run with a model size and quant, so there is no median to quote. The reports are listed below with how each one ran.
4 mlx-vlm reports on the M3 Ultra 512GB
Newest first. A run on several cards, with speculative decoding, with part of the model in system RAM or with batched requests says so, since each describes a different setup.
- Gemma 4 31B · 4bit MLX21.4 t/s
- Qwen3.5 397B · 17B active · 4bit MLX33.0 t/s
- Gemma 4 26B · 4B active · 4bit MLX67.6 t/s
- Qwen3.8 · BF16 unquantized13.2 t/s
Frequently asked
How fast is mlx-vlm on the M3 Ultra 512GB?
llamaperf has 4 mlx-vlm reports on the M3 Ultra 512GB, none of them a plain single-card run with a model size and quant, so there is no median to quote. The reports are listed below with how each one ran.