GPUs & Macs › AMD Radeon Pro VII
AMD Radeon Pro VII: local LLM speed
Llama 2 7B (Q4_0) generates 108 tokens/s on the AMD Radeon Pro VII (Vulkan), and processes prompts at 784 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 108 | 784 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 106 | 912 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 87.5 | 1,065 | Q4_0 | ROCm FA off | tg128 | llama.cpp #15021 details |
What fits on the AMD Radeon Pro VII
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 108 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| Qwen3.8-27B | UD-IQ3_XXS | 10.9 GB | 37.9 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 32.7 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 43.5 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 92.5 estimate |
| Qwen3.6-35B-A3B | UD-IQ3_XXS | 13.2 GB | 183 estimate |
| gemma-4-26B-A4B-it | UD-IQ4_XS | 13.6 GB | 99.1 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 43.4 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 50.6 estimate |
| Muse-Glimmer-30B | UD-IQ3_XXS | 13.1 GB | 31.6 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 96.9 estimate |
| gemma-4-31B-it | UD-IQ3_XXS | 11.8 GB | 35.0 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 61.7 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 346 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 83.5 estimate |
Similar hardware
AMD Radeon AI PRO R9700 · AMD Radeon Pro V620 · AMD Radeon Pro W7900