GPUs & Macs › AMD Instinct MI60
AMD Instinct MI60: local LLM speed
Llama 2 7B (Q4_0) generates 91.5 tokens/s on the AMD Instinct MI60 (ROCm), and processes prompts at 1,289 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 91.5 | 1,289 | Q4_0 | ROCm FA off | tg128 | llama.cpp #15021 details |
| Llama 2 7B | 78.2 | 369 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
What fits on the AMD Instinct MI60
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 91.5 tok/s (ROCm, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | Q8_0 | 29.0 GB | 12.0 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 27.6 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 36.7 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 78.0 estimate |
| Qwen3.6-35B-A3B | UD-Q6_K | 29.3 GB | 69.6 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 42.3 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 36.6 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 42.7 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 11.8 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 81.7 estimate |
| gemma-4-31B-it | Q6_K | 25.2 GB | 13.9 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 79.8 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 52.1 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 291 estimate |
Similar hardware
AMD Instinct MI50 · AMD Instinct MI100 · AMD Instinct MI25 · AMD Instinct MI210 · AMD Instinct MI300X