GPUs & Macs › AMD Radeon RX 6650 XT
AMD Radeon RX 6650 XT: local LLM speed
Llama 2 7B (Q4_0) generates 64.5 tokens/s on the AMD Radeon RX 6650 XT (Vulkan), and processes prompts at 1,089 tokens/s.
Measured results
| Model | Generate tok/s | Prompt tok/s | Quant | Backend | Test | Source |
|---|---|---|---|---|---|---|
| Llama 2 7B | 64.5 | 1,089 | Q4_0 | Vulkan FA on | tg128 | llama.cpp #10879 details |
| Llama 2 7B | 62.1 | 1,030 | Q4_0 | Vulkan FA off | tg128 | llama.cpp #10879 details |
| Llama 3.2 1B Instruct | 88.2 | 2,641 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
| Llama 3.1 8B Instruct | 25.2 | 536 | Q4_K_M | llamafile | LocalScore avg | LocalScore details |
What fits on the AMD Radeon RX 6650 XT
Estimates for this chip are scaled from a measured run: Llama 2 7B Q4_0 at 64.5 tok/s (Vulkan, source). Fit assumes the whole model in GPU memory with ~8K context.
| Model | Best quant that fits | File | Expected tok/s |
|---|---|---|---|
| gemma-4-12B-it | IQ4_XS | 6.4 GB | 38.7 estimate |
| Qwen3.5-9B | Q5_K_S | 6.4 GB | 38.8 estimate |
| Qwen3.5-4B | Q8_0 | 4.5 GB | 55.0 estimate |
| MiMo-V2.6-Distill-Qwen-9B | Q4_K_M | 5.8 GB | 42.2 estimate |
| gemma-4-E4B-it | Q5_K_M | 5.5 GB | 45.0 estimate |
| Qwen3-4B | Q8_0 | 4.3 GB | 57.6 estimate |
| gemma-3-1b-it | Q8_0 | 1.1 GB | 206 estimate |
| gemma-4-E2B-it | Q8_0 | 5.0 GB | 49.7 estimate |
| DeepSeek-R1-0528-Qwen3-8B | Q5_K_M | 5.9 GB | 42.2 estimate |
| Llama-3.2-3B-Instruct | Q8_0 | 3.4 GB | 72.1 estimate |
| Phi-4-mini-instruct | Q8_0 | 4.1 GB | 60.4 estimate |
| gemma-3-4b-it | Q8_0 | 4.1 GB | 59.7 estimate |
| Qwen3-14B | UD-IQ3_XXS | 6.0 GB | 41.0 estimate |
| gemma-3-12b-it | Q3_K_M | 6.0 GB | 41.1 estimate |
Similar hardware
AMD Radeon RX 5700 XT · AMD Radeon RX Vega 64 · AMD Radeon RX 7600 · AMD Radeon RX 6600 XT · AMD Radeon RX 6600 · AMD Radeon RX Vega 56 · AMD Radeon RX 580 · AMD Radeon RX 570