By memory size
| Memory | Run well | Biggest that runs well | gpt-oss-20b |
|---|---|---|---|
| 64 GB800 GB/s | 39 | Nemotron 3.5 Bolt 74B A9B | 189–337 tok/s |
| 128 GB800 GB/s | 47 | Qwen3.5 122B A10B | 189–337 tok/s |
| 192 GB800 GB/s | 49 | MiniMax M2.7 | 189–337 tok/s |
Models: Mac Pro (2023), Mac Pro (Rack, 2023). Not every memory size is offered with every model; Apple’s Tech Specs list what each one comes with.
Questions
- Is the Mac Pro with M2 Ultra good for local LLMs?
- With 192 GB it runs 49 of 73 open models well, up to MiniMax M2.7, and writes about 189–337 tokens a second with gpt-oss-20b. The M2 Ultra reads memory at 800 GB/s, which sets how fast it writes.
- How much memory should a Mac Pro with M2 Ultra have for AI?
- With 64 GB it runs 39 open models well; with 192 GB, 49, up to MiniMax M2.7. Memory decides which models fit, and it can’t be upgraded later.
Other Macs
Other Macs with the M2 Ultra:
Head to head:
The M2 Ultra for local AI · The best Mac for local LLMs · Every Mac
How it’s worked out
Memory: the model’s parameters at 8, 6, 4, 3 or 2 bits per weight plus MLX’s group scales (half a bit per weight), or its native format, with room for the context on top. By default macOS lets the GPU use about 75% of unified memory. Speed: memory bandwidth divided by the bytes read per token is the ceiling for one conversation; in llama.cpp’s public Apple Silicon benchmarks real runs reach about 45% to 80% of it, which is the range shown. These are estimates, not measurements: measure your own Mac on MLX Bench. Model figures from each official model card, chip figures from Apple’s Tech Specs. Checked 2026-10-19. Same data as mlx.app/compare.