Mac Studio M3 Ultra 512GB (Geerling test)
| GPU(s) | Apple M3 Ultra (unified) |
|---|---|
| VRAM / unified memory (GB) | 512 |
| CPU | Apple M3 Ultra |
| Motherboard | |
| RAM | |
| PSU | |
| Other | |
| Reported total cost | |
| Power | Peak 261.8 W (DeepSeek R1 671B), 243 W (Llama 3.1 70B), 227 W (gpt-oss 20B), 223 W (Llama 3.2 3B) |
Measured speed (generation tokens/sec)
📏 How measured: reported by the original source; RigBase re-checked the numbers at that source
| Model | Quant | Engine | tok/s | Details |
|---|---|---|---|---|
| DeepSeek R1 671B | llama.cpp | 19.89 | ||
| gpt-oss 20B | llama.cpp | 115.29 | ||
| gpt-oss 20B | MXFP4 | llama.cpp (Metal, llama-bench) | 115.52 | |
| Llama 3.1 70B | llama.cpp | 14.08 | ||
| Llama 3.1 405B | llama.cpp | 1.96 | ||
| DeepSeek R1 Distill 14B | llama.cpp | 51.85 | ||
| Llama 3.2 3B | llama.cpp | 154.6 |
Numbers from Jeff Geerling's ai-benchmarks results table (llama.cpp/Ollama eval rate). Quantization not stated in the summary table. Power is peak whole-system draw during that test. The 115.52 MXFP4 figure is from the llama.cpp gpt-oss guide.
Source: https://github.com/geerlingguy/ai-benchmarks/issues/42
https://github.com/ggml-org/llama.cpp/discussions/15396
https://github.com/geerlingguy/ai-benchmarks
Blank fields = not stated in the source. We never estimate build data; GPU street prices above are separately sourced and dated.
Rent before you buy: try this model on a cloud GPU for a few dollars first: Runpod (referral link; sign up with Google to get the new-user bonus)
Quick answers: Can Apple M3 Ultra run DeepSeek R1 671B? · Can Apple M3 Ultra run gpt-oss 20B?