GPUs & Macs › Apple M2 Ultra (76-core GPU)
Apple M2 Ultra (76-core GPU): local LLM speed
Llama 2 7B (Q4_0) generates 94.3 tokens/s on the Apple M2 Ultra (76-core GPU) (Metal), and processes prompts at 1,238 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 94.3 | 1,238 | Q4_0 | Metal FA off | tg128 | llama.cpp #4167 details |
| Llama 2 7B | 66.6 | 1,249 | Q8_0 | Metal FA off | tg128 | llama.cpp #4167 details |
| Llama 2 7B | 41.0 | 1,402 | F16 | Metal FA off | tg128 | llama.cpp #4167 details |
| gpt-oss 20B | 116 | 2,191 | MXFP4 | Metal FA on | tg128 | llama.cpp #15396 details |
| gpt-oss 120B | 79.7 | 1,245 | MXFP4 | Metal FA on | tg128 | llama.cpp #15396 details |
| Qwen2.5 14B Instruct | 36.6 | 470 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
What fits on the Apple M2 Ultra (76-core GPU)
Estimates for this chip are scaled from a measured run: Llama 2 7B F16 at 41.0 tok/s (Metal, source). Fit assumes the whole model in GPU memory with ~8K context.
With 16 GB unified memory
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | UD-IQ2_XXS | 7.3 GB | 76.1 estimate |
| gemma-4-12B-it | Q6_K | 9.8 GB | 56.5 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 58.0 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 123 estimate |
| Qwen3.6-35B-A3B | UD-IQ1_M | 10.0 GB | 321 estimate |
| gemma-4-26B-A4B-it | UD-IQ2_M | 10.0 GB | 179 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 57.9 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 67.5 estimate |
With 36 GB unified memory
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | UD-Q6_K | 22.0 GB | 25.1 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 43.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 58.0 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 123 estimate |
| Qwen3.6-35B-A3B | UD-Q5_K_S | 24.9 GB | 129 estimate |
| gemma-4-26B-A4B-it | UD-Q6_K | 23.2 GB | 77.6 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 57.9 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 67.5 estimate |
With 64 GB unified memory
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | Q8_0 | 29.0 GB | 19.0 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 43.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 58.0 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 123 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 87.4 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 66.9 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 57.9 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 67.5 estimate |
With 128 GB unified memory
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | Q8_0 | 29.0 GB | 19.0 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 43.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 58.0 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 123 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 87.4 estimate |
| Qwen3.8-Flash-Next | UD-IQ3_XXS | 82.0 GB | 6.7 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 66.9 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 57.9 estimate |
Similar hardware
Apple M1 (7-core GPU) · Apple M1 (8-core GPU) · Apple M1 Pro (14-core GPU) · Apple M1 Pro (16-core GPU) · Apple M1 Max (24-core GPU) · Apple M1 Max (32-core GPU) · Apple M1 Ultra (48-core GPU) · Apple M1 Ultra (64-core GPU)