← CanIRunLocal · All hardware · All models
Yes, if the Mac GPU limit is raised. Qwen 2.5 72B on M2 Max (38c GPU) 64GB at Q4_K_M needs sudo sysctl iogpu.wired_limit_mb=54375 (51.1 GB needed, 48.0 GB usable), about 7 tok/s (estimate).
Hardware: M2 Max (38c GPU) 64GB · Model: Qwen 2.5 72B
M2 Max (38c GPU) 64GB has 64 GB of unified memory and 400 GB/s of memory bandwidth. M1 Max (32c GPU) 64GB and M2 Max (30c GPU) 64GB and M3 Max (40c GPU) 64GB have the same memory and bandwidth, so the fit numbers match; this page is M2 Max (38c GPU) 64GB. The checker leaves macOS a share of unified memory, so Ollama can use 48.0 GB. Qwen 2.5 72B (72.7B, Alibaba) at Q4_K_M is a 47 GB download and needs 51.1 GB with headroom for about 8k tokens.
| Best quant | Q4_K_M |
|---|---|
| Download size | 47 GB |
| Memory needed | 51.1 GB |
| Usable fast memory | 48.0 GB |
| Fit | fits if the Mac GPU limit is raised |
| Estimated speed | ~7 tok/s (estimate) |
| Mac GPU limit | sudo sysctl iogpu.wired_limit_mb=54375 |
| Command | ollama run qwen2.5:72b |
Estimates, not benchmarks. Model sizes are ollama.com download sizes; memory needed adds headroom for ~8k tokens of context; speeds are rough figures from memory bandwidth. Data updated 2026-10-23.
A project by OnDevice Labs · llms.txt