GPUs & Macs › AMD Radeon Pro W7900
AMD Radeon Pro W7900: local LLM speed
Llama 2 7B (Q4_0) generates 127 tokens/s on the AMD Radeon Pro W7900 (ROCm), and processes prompts at 3,473 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 127 | 3,473 | Q4_0 | ROCm FA on | tg128 | llama.cpp #15021 details |
| Llama 2 7B | 121 | 3,213 | Q4_0 | ROCm FA off | tg128 | llama.cpp #15021 details |
What fits on the AMD Radeon Pro W7900
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 127 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 | 16.8 estimate |
| gemma-4-12B-it | Q8_0 | 12.7 GB | 38.4 estimate |
| Qwen3.5-9B | Q8_0 | 9.5 GB | 51.1 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 109 estimate |
| Qwen3.6-35B-A3B | Q8_0 | 36.9 GB | 77.0 estimate |
| gemma-4-26B-A4B-it | Q8_0 | 26.9 GB | 59.0 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q8_0 | 9.5 GB | 51.0 estimate |
| gemma-4-E4B-it | Q8_0 | 8.2 GB | 59.5 estimate |
| Muse-Glimmer-30B | Q8_0 | 29.6 GB | 16.4 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 114 estimate |
| gemma-4-31B-it | Q8_0 | 32.6 GB | 14.9 estimate |
| gemma-4-26B-A4B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 14.2 GB | 111 estimate |
| gemma-4-12B-it-qat-q4_0-unquantized | UD-Q4_K_XL | 6.7 GB | 72.5 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 406 estimate |
Similar hardware
AMD Radeon AI PRO R9700 · AMD Radeon Pro V620 · AMD Radeon Pro VII