2x AMD Strix Halo
其他45
运行记录
15
覆盖模型
4
覆盖框架
L2 跨框架验证 × 45
证据分布
典型性能表现
同硬件上各模型 × 框架 × 量化的已上架实测均值。
- llama.cpp · FP16 Decode 94.9 · Prefill 326.17 ·
- Strata · Q5_K_M Decode 206.52 · Prefill 718.15 ·
- Ollama · FP16 Decode 58.36 · Prefill 233.8 ·
- Strata · IQ3_XS Decode 234.55 · Prefill 886.87 ·
- Ollama · IQ2_M Decode 146.37 · Prefill 764.74 ·
- Strata · IQ2_XS Decode 213.76 · Prefill 822.5 ·
- vLLM · FP16 Decode 163.9 · Prefill 573.99 ·
- vLLM · IQ3_XS Decode 352.1 · Prefill 1673.75 ·
- Strata · Q2_K Decode 284.75 · Prefill 1118.17 ·
- Strata · Q5_K_S Decode 358.45 · Prefill 1286.06 ·
- llama.cpp · IQ1_S Decode 185.41 · Prefill 664.26 ·
- llama.cpp · IQ4_XS Decode 145.51 · Prefill 650.25 ·
- Strata · IQ2_M Decode 243.64 · Prefill 1152.92 ·
- Strata · IQ3_S Decode 220.16 · Prefill 1160.56 ·
- vLLM · Q3_K_L Decode 159.39 · Prefill 715.55 ·
- vLLM · Q4_K_S Decode 137.45 · Prefill 721.98 ·
- Strata · IQ1_S Decode 293.93 · Prefill 1381.03 ·
- Strata · Q5_K_M Decode 157.71 · Prefill 787.63 ·
- llama.cpp · FP16 Decode 87.02 · Prefill 443.84 ·
- Ollama · IQ3_S Decode 197.15 · Prefill 955.03 ·
- Ollama · Q4_K_M Decode 165.67 · Prefill 892.04 ·
- llama.cpp · Q5_K_S Decode 236.18 · Prefill 1075.25 ·
- Strata · Q5_K_M Decode 284.45 · Prefill 1280.23 ·
- Strata · Q5_K_S Decode 326.25 · Prefill 1416.72 ·
- llama.cpp · Q3_K_M Decode 181.31 · Prefill 944.53 ·
- llama.cpp · Q4_K_M Decode 153.87 · Prefill 676.15 ·
- Strata · IQ2_M Decode 312.05 · Prefill 1568.66 ·
- vLLM · Q3_K_S Decode 149.19 · Prefill 617.1 ·
- vLLM · Q8_0 Decode 87.69 · Prefill 408.22 ·
- Strata · Q3_K_M Decode 208.39 · Prefill 864.96 ·
- Ollama · Q3_K_S Decode 347.47 · Prefill 1631.19 ·
- Ollama · Q8_0 Decode 256.99 · Prefill 1220.11 ·
- Strata · Q4_K_M Decode 303.04 · Prefill 1091.32 ·
- llama.cpp · IQ2_XS Decode 273.09 · Prefill 1000.97 ·
- llama.cpp · IQ3_S Decode 239.07 · Prefill 794.3 ·
- vLLM · Q5_K_M Decode 183.47 · Prefill 864.96 ·
- vLLM · Q6_K Decode 171.22 · Prefill 756.33 ·
- Ollama · FP16 Decode 181.91 · Prefill 736.26 ·
- Ollama · IQ4_XS Decode 379.81 · Prefill 1812.54 ·
- Strata · IQ4_XS Decode 288.7 · Prefill 1574.67 ·
- vLLM · Q3_K_S Decode 277.77 · Prefill 1024.78 ·
- Strata · IQ3_XS Decode 321.59 · Prefill 1311.64 ·
- vLLM · IQ3_S Decode 197.35 · Prefill 823.81 ·
- vLLM · Q5_K_M Decode 149.23 · Prefill 534.97 ·
- Strata · Q2_K Decode 311.82 · Prefill 1554.79 ·
Fit 判定
按显存与内存带宽做物理估算:选择模型,查看各量化能否装下及预计速度
该硬件未收录显存数据,无法判定 Fit
估算公式:tok/s =(带宽 ÷ 模型占用)× 0.55,借鉴 llmfit;实测以运行记录为准
核心规格
- 厂商
- AMD
- 显存带宽
- 512 GB/s
运行记录
共 45 组配置
| 模型 | 框架 | 量化 | Decode | 生成耗时 | 样本数 |
|---|---|---|---|---|---|
| Qwen3.5-4B | llama.cpp | FP16 | 94.9 tok/s | — | 1 |
| Qwen3.5-4B | Strata | Q5_K_M | 206.5 tok/s | — | 1 |
| Qwen3.5-9B | Ollama | FP16 | 58.4 tok/s | — | 1 |
| Qwen3.5-9B | Strata | IQ3_XS | 234.6 tok/s | — | 1 |
| Qwen3.5-27B | Ollama | IQ2_M | 146.4 tok/s | — | 1 |
| Qwen3.5-27B | Strata | IQ2_XS | 213.8 tok/s | — | 1 |
| Qwen3.5-0.8B | vLLM | FP16 | 163.9 tok/s | — | 1 |
| Qwen3.5-0.8B | vLLM | IQ3_XS | 352.1 tok/s | — | 1 |
| Qwen3.5-0.8B | Strata | Q2_K | 284.8 tok/s | — | 1 |
| Qwen3.5-0.8B | Strata | Q5_K_S | 358.4 tok/s | — | 1 |
| Qwen3.8-9B | llama.cpp | IQ1_S | 185.4 tok/s | — | 1 |
| Qwen3.8-9B | llama.cpp | IQ4_XS | 145.5 tok/s | — | 1 |
| Qwen3.8-9B | Strata | IQ2_M | 243.6 tok/s | — | 1 |
| Qwen3.8-9B | Strata | IQ3_S | 220.2 tok/s | — | 1 |
| Qwen3-8B | vLLM | Q3_K_L | 159.4 tok/s | — | 1 |
| Qwen3-8B | vLLM | Q4_K_S | 137.4 tok/s | — | 1 |
| Qwen3-8B | Strata | IQ1_S | 293.9 tok/s | — | 1 |
| Qwen3-8B | Strata | Q5_K_M | 157.7 tok/s | — | 1 |
| Qwen3-4B | llama.cpp | FP16 | 87 tok/s | — | 1 |
| Qwen3-4B | Ollama | IQ3_S | 197.2 tok/s | — | 1 |
| Qwen3-4B | Ollama | Q4_K_M | 165.7 tok/s | — | 1 |
| Qwen3-1.7B | llama.cpp | Q5_K_S | 236.2 tok/s | — | 1 |
| Qwen3-1.7B | Strata | Q5_K_M | 284.4 tok/s | — | 1 |
| Qwen3-1.7B | Strata | Q5_K_S | 326.3 tok/s | — | 1 |
| Qwen3-4B-Thinking-2507 | llama.cpp | Q3_K_M | 181.3 tok/s | — | 1 |
| Qwen3-4B-Thinking-2507 | llama.cpp | Q4_K_M | 153.9 tok/s | — | 1 |
| Qwen3-4B-Thinking-2507 | Strata | IQ2_M | 312.1 tok/s | — | 1 |
| Qwen3-Coder-Next | vLLM | Q3_K_S | 149.2 tok/s | — | 1 |
| Qwen3-Coder-Next | vLLM | Q8_0 | 87.7 tok/s | — | 1 |
| Qwen3-Coder-Next | Strata | Q3_K_M | 208.4 tok/s | — | 1 |
| Qwen2.5-0.5B-Instruct | Ollama | Q3_K_S | 347.5 tok/s | — | 1 |
| Qwen2.5-0.5B-Instruct | Ollama | Q8_0 | 257 tok/s | — | 1 |
| Qwen2.5-0.5B-Instruct | Strata | Q4_K_M | 303 tok/s | — | 1 |
| Qwen2.5-Omni-3B | llama.cpp | IQ2_XS | 273.1 tok/s | — | 1 |
| Qwen2.5-Omni-3B | llama.cpp | IQ3_S | 239.1 tok/s | — | 1 |
| Qwen2.5-Omni-3B | vLLM | Q5_K_M | 183.5 tok/s | — | 1 |
| Qwen2.5-Omni-3B | vLLM | Q6_K | 171.2 tok/s | — | 1 |
| Qwen3-Embedding-0.6B | Ollama | FP16 | 181.9 tok/s | — | 1 |
| Qwen3-Embedding-0.6B | Ollama | IQ4_XS | 379.8 tok/s | — | 1 |
| Qwen3-Embedding-0.6B | Strata | IQ4_XS | 288.7 tok/s | — | 1 |
| gemma-4-E2B-it | vLLM | Q3_K_S | 277.8 tok/s | — | 1 |
| gemma-4-E2B-it | Strata | IQ3_XS | 321.6 tok/s | — | 1 |
| gemma-4-E4B-it | vLLM | IQ3_S | 197.3 tok/s | — | 1 |
| gemma-4-E4B-it | vLLM | Q5_K_M | 149.2 tok/s | — | 1 |
| gemma-4-E4B-it | Strata | Q2_K | 311.8 tok/s | — | 1 |