Mac mini M4 Pro
M4 Pro (20-core GPU) · 273 GB/s
Mac · Apple Silicon
M4 Pro (20-core GPU) with 48 GB: which local LLMs run?
With 48 GB of unified memory, the M4 Pro (20-core GPU) gives the GPU about 36 GB by default (273 GB/s): enough for models up to about Mixtral 8x7B Instruct.
- Unified memory48 GB
- GPU usable~36 GB
- Bandwidth273 GB/s
- GPU cores20
Unified memory is not VRAM: macOS reserves part of it. The limit can be raised with sudo sysctl iogpu.wired_limit_mb=…, at the cost of memory for apps. There is no offload to system RAM — it is the same memory.
- CPU cores
- 14
- GPU cores
- 20
- Bandwidth
- 273 GB/s
- Memory configurations
- 24 GB · 48 GB · 64 GB
- Sold in
- Mac mini, MacBook Pro 14, MacBook Pro 16
- Released
- October 2024
Official specData updated 24 days agoSource: Apple tech specs
What the M4 Pro (20-core GPU) with 48 GB runs
| gpt-oss-20bMXFP4 · 8K~12.8 GB · 31–70 tok/s estimate | Excellent |
|---|---|
| Qwen3 14BQ8_0 · 8K~17.3 GB · 11–16 tok/s estimate | Excellent |
| DeepSeek R1 Distill Qwen 14BQ8_0 · 8K~17.6 GB · 11–16 tok/s estimate | Excellent |
| Qwen2.5 14B InstructQ8_0 · 8K~17.6 GB · 11–16 tok/s estimate | Excellent |
| Phi-4 14BQ8_0 · 8K~17.4 GB · 11–16 tok/s estimate | Excellent |
| Mistral Nemo 12B InstructQ8_0 · 8K~14.7 GB · 13–19 tok/s estimate | Excellent |
| Gemma 3 12B ITQ8_0 · 8K~13.9 GB · 13–20 tok/s estimate | Excellent |
| Gemma 2 9B ITQ8_0 · 8K~12.5 GB · 16–23 tok/s estimate | Excellent |
| Qwen3 8BQ8_0 · 8K~10.3 GB · 18–27 tok/s estimate | Excellent |
| Llama 3.1 8B InstructQ8_0 · 8K~10 GB · 19–28 tok/s estimate | Excellent |
| DeepSeek R1 Distill Llama 8BQ8_0 · 8K~10 GB · 19–28 tok/s estimate | Excellent |
| Qwen2.5 7B InstructQ8_0 · 8K~9 GB · 21–30 tok/s estimate | Excellent |
| Mistral 7B Instruct v0.3Q8_0 · 8K~9 GB · 21–31 tok/s estimate | Excellent |
| Gemma 3 4B ITQ8_0 · 8K~5.2 GB · 37–59 tok/s estimate | Excellent |
| Qwen3 4BQ8_0 · 8K~6 GB · 33–52 tok/s estimate | Excellent |
| Phi-4 Mini Instruct 3.8BQ8_0 · 8K~5.8 GB · 35–55 tok/s estimate | Excellent |
| Llama 3.2 3B InstructQ8_0 · 8K~4.9 GB · 41–65 tok/s estimate | Excellent |
| Qwen3 1.7BQ8_0 · 8K~3.5 GB · 65–109 tok/s estimate | Excellent |
| Llama 3.2 1B InstructQ8_0 · 8K~2.3 GB · 104–173 tok/s estimate | Excellent |
| Qwen3 0.6BQ8_0 · 8K~2.3 GB · 112–227 tok/s estimate | Excellent |
| Mixtral 8x7B InstructQ5_K_M · 8K~33 GB · 15–22 tok/s estimate | Compatible |
| Qwen3 32BQ6_K · 8K~29 GB · 6–9 tok/s estimate | Compatible |
| DeepSeek R1 Distill Qwen 32BQ6_K · 8K~29 GB · 6–9 tok/s estimate | Compatible |
| Qwen2.5 32B InstructQ6_K · 8K~29 GB · 6–9 tok/s estimate | Compatible |
| Qwen2.5 Coder 32B InstructQ6_K · 8K~29 GB · 6–9 tok/s estimate | Compatible |
| QwQ 32BQ6_K · 8K~29 GB · 6–9 tok/s estimate | Compatible |
| Qwen3 30B-A3BQ8_0 · 8K~33.2 GB · 28–43 tok/s estimate | Compatible |
| Qwen3 Coder 30B-A3B InstructQ8_0 · 8K~33.2 GB · 28–43 tok/s estimate | Compatible |
| Gemma 3 27B ITQ8_0 · 8K~30 GB · 6–8.5 tok/s estimate | Compatible |
| Gemma 2 27B ITQ8_0 · 8K~31.4 GB · 6–8.5 tok/s estimate | Compatible |
| Mistral Small 3.2 24B InstructQ8_0 · 8K~26.4 GB · 7–10 tok/s estimate | Compatible |
| Qwen2.5 72B InstructQ3_K_M · 8K~38 GB · No data | Compatible with limits |
| Llama 3.3 70B InstructQ3_K_M · 8K~36.8 GB · No data | Compatible with limits |
| DeepSeek R1 Distill Llama 70BQ3_K_M · 8K~36.8 GB · No data | Compatible with limits |
| DeepSeek R1 (671B)Q8_0 · 8K~699 GB · No data | Incompatible |
| Qwen3 235B-A22BQ8_0 · 8K~247 GB · No data | Incompatible |
| gpt-oss-120bMXFP4 · 8K~63 GB · No data | Incompatible |
| Llama 4 Scout 17B-16E InstructQ8_0 · 8K~113 GB · No data | Incompatible |
For each model, the highest-precision quantization that runs entirely in fast memory; otherwise the one with the best result. Speeds are ranges; "estimate" unless a benchmark exists for this exact hardware.
EstimateData updated 24 days agoSource: Run LLM Locally engine
Where to buy
Prices and links follow the market you choose.
No current offer imported for this market: the link opens a marketplace search.