What fits in 512 GB of unified memory
Data updated . Last reviewed the same day, when this catalog snapshot was written.
2 catalog machines have 512 GB unified. Models are ranked against the Mac Studio M5 Ultra 512GB, the tightest usable pool at this sticker size (358 GB). The tok/s estimate uses Mac Studio M5 Ultra 512GB bandwidth (1200 GB/s) and is not a benchmark. 3 of these fits do not clear the 128 GB tier.
Models
| Model | Best quant | Estimate | Headroom | Est. tok/s |
|---|---|---|---|---|
| Llama 3.2 1B Instruct | FP16 | 3.0 GB | 355 GB | 485 tok/s |
| Llama 3.2 3B Instruct | FP16 | 7.7 GB | 351 GB | 187 tok/s |
| Phi-3.5 mini | FP16 | 9.2 GB | 349 GB | 157 tok/s |
| Phi-4 mini | FP16 | 9.2 GB | 349 GB | 156 tok/s |
| Gemma 3 4B | FP16 | 10.3 GB | 348 GB | 140 tok/s |
| Mistral 7B Instruct v0.3 | FP16 | 17.4 GB | 341 GB | 82.8 tok/s |
| Qwen2.5 7B Instruct | FP16 | 18.3 GB | 340 GB | 78.8 tok/s |
| Gemma 4 E4B | FP16 | 19.2 GB | 339 GB | 75.0 tok/s |
| DeepSeek R1 Distill Llama 8B | FP16 | 19.3 GB | 339 GB | 74.7 tok/s |
| Llama 3.1 8B Instruct | FP16 | 19.3 GB | 339 GB | 74.7 tok/s |
| Qwen3 8B | FP16 | 19.7 GB | 339 GB | 73.3 tok/s |
| Ministral 3 8B | FP16 | 21.4 GB | 337 GB | 67.3 tok/s |
| Gemma 2 9B | FP16 | 22.2 GB | 336 GB | 64.9 tok/s |
| Gemma 4 12B | FP16 | 28.7 GB | 330 GB | 50.2 tok/s |
| Gemma 3 12B | FP16 | 29.2 GB | 329 GB | 49.2 tok/s |
| Mistral Nemo 12B | FP16 | 29.4 GB | 329 GB | 49.0 tok/s |
| Ministral 3 14B | FP16 | 33.5 GB | 325 GB | 43.0 tok/s |
| Phi-4 | FP16 | 35.2 GB | 323 GB | 40.9 tok/s |
| Qwen3 14B | FP16 | 35.4 GB | 323 GB | 40.6 tok/s |
| Qwen2.5 14B Instruct | FP16 | 35.4 GB | 323 GB | 40.6 tok/s |
| gpt-oss-20b | FP16 | 50.2 GB | 308 GB | 167 tok/s |
| Mistral Small 3.2 24B | FP16 | 57.6 GB | 301 GB | 25.0 tok/s |
| Gemma 4 26B-A4B | FP16 | 61.9 GB | 296 GB | 158 tok/s |
| Gemma 2 27B | FP16 | 65.3 GB | 293 GB | 22.0 tok/s |
| Gemma 3 27B | FP16 | 65.8 GB | 293 GB | 21.9 tok/s |
| Qwen3.6 27B | FP16 | 66.7 GB | 292 GB | 21.6 tok/s |
| Qwen3.8 27B | FP16 | 66.7 GB | 292 GB | 21.6 tok/s |
| Qwen3 30B-A3B | FP16 | 73.3 GB | 285 GB | 200 tok/s |
| GLM-4.7-Flash | FP16 | 74.9 GB | 283 GB | 200 tok/s |
| Gemma 4 31B | FP16 | 75.1 GB | 283 GB | 19.2 tok/s |
| Qwen3 32B | FP16 | 78.6 GB | 280 GB | 18.3 tok/s |
| DeepSeek R1 Distill Qwen 32B | FP16 | 78.6 GB | 280 GB | 18.3 tok/s |
| Qwen2.5 32B Instruct | FP16 | 78.6 GB | 280 GB | 18.3 tok/s |
| Qwen3.6 35B-A3B | FP16 | 86.3 GB | 272 GB | 200 tok/s |
| Mixtral 8x7B | FP16 | 112 GB | 246 GB | 46.5 tok/s |
| DeepSeek R1 Distill Llama 70B | FP16 | 169 GB | 189 GB | 8.50 tok/s |
| Llama 3.3 70B Instruct | FP16 | 169 GB | 189 GB | 8.50 tok/s |
| Qwen2.5 72B Instruct | FP16 | 174 GB | 184 GB | 8.25 tok/s |
| Llama 4 Scout | FP16 | 261 GB | 97.7 GB | 35.3 tok/s |
| gpt-oss-120b | FP16 | 280 GB | 78.0 GB | 118 tok/s |
| Mistral Small 4 | FP16 | 287 GB | 71.8 GB | 92.3 tok/s |
| Mixtral 8x22B | FP16 | 338 GB | 20.9 GB | 15.4 tok/s |
| DeepSeek V4 Flash | Q8 | 4.5 GB | 354 GB | 7053 tok/s |
| Qwen3 235B-A22B | Q8 | 252 GB | 107 GB | 51.3 tok/s |
| Llama 4 Maverick | Q4 | 243 GB | 116 GB | 117 tok/s |
| Llama 3.1 405B Instruct | Q4 | 244 GB | 115 GB | 5.91 tok/s |
Machines at 512 GB
- Mac Studio M5 Ultra 512GB
358 GB usable · Price varies
- Mac Studio M3 Ultra 512GB
358 GB usable · Price varies
Questions
- How many models fit in 512 GB of unified memory?
- 46 of 49 fit on the tightest config at this size, the Mac Studio M5 Ultra 512GB, which keeps 358 GB usable.
- Which catalog machines have 512 GB unified?
- Mac Studio M5 Ultra 512GB (70% usable), Mac Studio M3 Ultra 512GB (70% usable).
- Why is the ranking based on the tightest config?
- So a model listed as fitting works on every 512 GB machine here, including ones that reserve more memory for the OS. A looser machine has more headroom than the table shows.