GPUs & Macs › AMD Radeon Pro V620
AMD Radeon Pro V620: local LLM speed
Llama 2 7B (Q4_0) generates 81.8 tokens/s on the AMD Radeon Pro V620 (Vulkan), and processes prompts at 1,595 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 81.8 | 1,595 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 79.2 | 1,556 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 74.7 | 1,804 | Q4_0 | ROCm FA off | tg128 | llama.cpp #15021 details |
| Llama 2 7B | 70.8 | 1,257 | Q4_0 | ROCm FA on | tg128 | llama.cpp #15021 details |
| Llama 3.2 1B Instruct | 57.0 | 3,905 | Q4_K_M | llamafile | LocalScore avg · n=2 | LocalScore LocalScore details |
What fits on the AMD Radeon Pro V620
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 81.8 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 | Q8_0 | 29.0 GB | 10.8 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 24.7 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 32.8 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 69.7 estimate |
| Qwen3.6-35B-A3B | UD-Q6_K | 29.3 GB | 62.2 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 37.8 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 32.7 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 38.2 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 10.6 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 73.0 estimate |
| gemma-4-31B-it | Q6_K | 25.2 GB | 12.4 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 71.3 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 46.5 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 261 estimate |
Similar hardware
AMD Radeon AI PRO R9700 · AMD Radeon Pro VII · AMD Radeon Pro W7900