趋近智
AMD Ryzen AI 9 HX 370 (96GB system, 72GB GPU-accessible)
AMD Ryzen AI 9 HX 370 (96GB system, 72GB GPU-accessible) 提供 72 GB 显存/内存与 128 GB/s 带宽，支持通过 ROCm 与 Vulkan 加速进行本地推理。
总显存
72 GB
可用显存上限
72 GB
显存带宽
128 GB/s
最大推理模型级别
70B+ 模型 (Q4)
热设计功耗
54W
显存占用随上下文长度变化曲线
随上下文长度增长的显存需求评估。当曲线超过参考线时将导致显存不足（OOM）。
量化精度:
AMD Ryzen AI 9 HX 370 (96GB system, 72GB GPU-accessible) 最佳运行大模型与速度矩阵
主流开源大模型在 AMD Ryzen AI 9 HX 370 (96GB system, 72GB GPU-accessible) 上的推理速度、显存兼容性与生成吞吐表现。
排序:
上下文长度:
KV 缓存:
显卡数量:
| 排名 | 模型 | 参数量 | 量化精度 | 预估 TPS | TTFT | 可运行 |
|---|---|---|---|---|---|---|
#36 | Qwen 3.8 27B | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
#69 | Sarvam-105B | 106B (10.3B active) | INT4 | 17 tok/s | ~12.6s | |
#80 | K2 Horizon MoVA 36B A4B | 36B (4B active) | INT4 Q8 FP16 | 33 tok/s 21 tok/s 11 tok/s | ~4.8s ~4.8s ~5.7s | |
#83 | Ling 3.0 Flash VL | 124B (5.5B active) | INT4 | 27 tok/s | ~9.7s | |
#86 | Qwen3.5-27B | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
#88 | Qwen3.5-122B-A10B | 122B (10B active) | INT4 | 12 tok/s | ~14.0s | |
#93 | Sarvam-30B | 32B (2.4B active) | INT4 Q8 FP16 | 46 tok/s 32 tok/s 18 tok/s | ~3.3s ~3.4s ~3.7s | |
#99 | Muse Glimmer 30B | 30B | INT4 Q8 FP16 | 6 tok/s 4 tok/s 2 tok/s | ~27.3s ~27.4s ~30.0s | |
#102 | Phi-4 Reasoning Plus | 14B | INT4 Q8 FP16 | 9 tok/s 6 tok/s 3 tok/s | ~12.9s ~13.0s ~13.1s | |
#103 | Qwen3.6 35B A3B | 35B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 14 tok/s | ~3.9s ~3.9s ~4.6s | |
#105 | MiMo V2 Flash | 15B (309B active) | INT4 Q8 FP16 | 1 tok/s 1 tok/s 1 tok/s | ~268.9s ~270.2s ~273.0s | |
#110 | Gemma 4 26B A4B | 25.2B (3.8B active) | INT4 Q8 FP16 | 36 tok/s 23 tok/s 13 tok/s | ~4.4s ~4.4s ~4.4s | |
#113 | GLM-4.5-Air | 106B (12B active) | INT4 | 4 tok/s | ~16.7s | |
#117 | Nemotron 3.5 Bolt 74B A9B | 74B | INT4 | 3 tok/s | ~66.7s | |
#120 | Qwen3.5-35B-A3B | 35B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 14 tok/s | ~3.9s ~3.9s ~4.6s | |
#121 | Qwen3.5-9B | 9B | INT4 Q8 FP16 | 20 tok/s 11 tok/s 6 tok/s | ~8.4s ~8.4s ~8.5s | |
#122 | Ling 3.0 Flash | 124B (5.1B active) | INT4 | 29 tok/s | ~9.3s | |
#127 | K2 Horizon 7B | 7B | INT4 Q8 FP16 | 25 tok/s 14 tok/s 7 tok/s | ~6.5s ~6.6s ~6.6s | |
#130 | Gemma 4 12B | 11.95B | INT4 Q8 FP16 | 11 tok/s 7 tok/s 4 tok/s | ~11.1s ~11.1s ~11.2s | |
#131 | Qwen3 Next 80B A3B | 80B (79B active) | INT4 | 2 tok/s | ~71.3s | |
#136 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 40 tok/s 24 tok/s 13 tok/s | ~3.8s ~3.9s ~3.9s | |
#139 | GPT-OSS 20B | 21B (3.6B active) | INT4 Q8 FP16 | 15 tok/s 13 tok/s 9 tok/s | ~4.1s ~4.2s ~4.2s | |
#140 | GLM-4V | 9B | INT4 Q8 FP16 | 14 tok/s 9 tok/s 5 tok/s | ~8.4s ~8.4s ~8.5s | |
#145 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~26.4s ~26.5s ~26.8s | |
#147 | Nemotron 3.5 Super VL | 120B | INT4 | 2 tok/s | ~117.7s | |
#150 | Qwen3-30B-A3B | 30B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 15 tok/s | ~3.8s ~3.8s ~4.2s | |
#152 | Granite 4.2 8B | 8B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#157 | Gemma 4 E4B | 8B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#158 | Mistral Large 3 | 41B (675B active) | INT4 Q8 | 1 tok/s 1 tok/s | ~587.0s ~589.8s | |
#159 | Qwen2.5-72B | 72B | INT4 Q8 | 3 tok/s 1 tok/s | ~64.9s ~76.6s | |
#161 | GLM-4 | 32B | INT4 Q8 FP16 | 5 tok/s 3 tok/s 1 tok/s | ~29.1s ~29.2s ~34.7s | |
#162 | DeepSeek-R1 70B | 70B | INT4 Q8 | 3 tok/s 1 tok/s | ~63.0s ~74.5s | |
#163 | Gemma 3 27B | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
#164 | Phi-4 | 14B | INT4 Q8 FP16 | 13 tok/s 7 tok/s 4 tok/s | ~12.9s ~13.0s ~13.1s | |
#165 | Llama 3.3 70B | 70B | INT4 Q8 | 3 tok/s 1 tok/s | ~63.1s ~74.5s | |
#166 | Qwen3-32B | 32B | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~29.0s ~29.2s ~32.0s | |
#167 | Hunyuan Standard | 52B (389B active) | INT4 Q8 | 1 tok/s 1 tok/s | ~339.1s ~370.2s | |
#168 | DeepSeek-R1 32B | 32B | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~29.0s ~29.2s ~32.0s | |
#169 | Gemma 4 31B | 30.7B | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~27.9s ~28.0s ~30.7s | |
#171 | Magistral Small | 24B | INT4 Q8 FP16 | 6 tok/s 4 tok/s 2 tok/s | ~21.9s ~22.0s ~22.3s | |
#172 | Qwen2.5-32B | 32B | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~29.0s ~29.2s ~32.0s | |
#174 | OLMo 3.1 32B Instruct | 32B | INT4 Q8 FP16 | 5 tok/s 3 tok/s 1 tok/s | ~29.1s ~29.2s ~34.7s | |
#175 | Llama 3.1 70B | 70B | INT4 Q8 | 3 tok/s 1 tok/s | ~63.1s ~74.5s | |
#176 | DeepSeek-R1 14B | 14B | INT4 Q8 FP16 | 13 tok/s 8 tok/s 4 tok/s | ~12.9s ~13.0s ~13.1s | |
#177 | Ministral 3 14B | 14B | INT4 Q8 FP16 | 9 tok/s 6 tok/s 3 tok/s | ~12.9s ~13.0s ~13.1s | |
#178 | Mistral-Large-2407 | 123B | INT4 | 1 tok/s | ~120.5s | |
#179 | Llama 4 Scout | 109B (17B active) | INT4 | 9 tok/s | ~20.1s | |
#180 | Gemma 3 12B | 12B | INT4 Q8 FP16 | 15 tok/s 9 tok/s 4 tok/s | ~11.1s ~11.2s ~11.3s | |
#182 | Command A | 111B | INT4 | 1 tok/s | ~117.9s | |
#183 | Qwen2-72B | 72B | INT4 Q8 | 3 tok/s 1 tok/s | ~64.9s ~76.6s | |
#184 | Mistral-Small-2501 | 24B | INT4 Q8 FP16 | 8 tok/s 4 tok/s 2 tok/s | ~21.9s ~22.0s ~22.2s | |
#185 | Ministral 3 8B | 8B | INT4 Q8 FP16 | 15 tok/s 10 tok/s 6 tok/s | ~7.5s ~7.5s ~7.6s | |
#186 | Gemma 2 27B | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
#187 | OLMo 3 7B Think | 7B | INT4 Q8 FP16 | 17 tok/s 12 tok/s 7 tok/s | ~6.6s ~6.6s ~6.7s | |
#188 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 70 tok/s 44 tok/s 24 tok/s | ~2.0s ~2.0s ~2.1s | |
#189 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 70 tok/s 44 tok/s 24 tok/s | ~2.0s ~2.0s ~2.1s | |
#190 | Qwen2.5-14B | 14B | INT4 Q8 FP16 | 13 tok/s 7 tok/s 4 tok/s | ~12.9s ~13.0s ~13.1s | |
#191 | OLMo 3.1 32B Think | 32B | INT4 Q8 FP16 | 5 tok/s 3 tok/s 1 tok/s | ~29.1s ~29.2s ~34.7s | |
#192 | Llama 3 70B | 70B | INT4 Q8 | 3 tok/s 1 tok/s | ~63.1s ~74.5s | |
#193 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 33 tok/s 19 tok/s 10 tok/s | ~4.8s ~4.8s ~4.9s | |
#194 | Command R Plus | 104B | INT4 | 1 tok/s | ~110.4s | |
#195 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 31 tok/s 23 tok/s 14 tok/s | ~2.9s ~3.0s ~3.0s | |
#196 | DeepSeek-R1 8B | 8B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#198 | Gemma 2 9B | 9B | INT4 Q8 FP16 | 20 tok/s 11 tok/s 6 tok/s | ~8.4s ~8.4s ~8.5s | |
#199 | Nemotron 3.5 Lightning 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 15 tok/s | ~3.8s ~3.8s ~4.2s | |
#200 | Kimi Linear 48B A3B Instruct | 48B (3B active) | INT4 Q8 | 11 tok/s 11 tok/s | ~4.2s ~4.2s | |
#201 | Qwen3-4B | 4B | INT4 Q8 FP16 | 40 tok/s 24 tok/s 13 tok/s | ~3.8s ~3.9s ~3.9s | |
#202 | Qwen3-14B | 14B | INT4 Q8 FP16 | 13 tok/s 7 tok/s 4 tok/s | ~12.9s ~13.0s ~13.1s | |
#203 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 40 tok/s 24 tok/s 13 tok/s | ~3.8s ~3.9s ~3.9s | |
#204 | DeepSeek-R1 7B | 7B | INT4 Q8 FP16 | 26 tok/s 15 tok/s 8 tok/s | ~6.5s ~6.6s ~6.6s | |
#205 | OLMo 3 32B Base | 32B | INT4 Q8 FP16 | 5 tok/s 3 tok/s 1 tok/s | ~29.1s ~29.2s ~34.7s | |
#206 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 135 tok/s 93 tok/s 55 tok/s | ~825 ms ~829 ms ~836 ms | |
#207 | Llama 3.1 8B | 8B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#207 | Qwen3-8B | 8B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#210 | Nemotron 3.5 Lightning | 30B (3B active) | INT4 Q8 FP16 | 14 tok/s 12 tok/s 9 tok/s | ~3.9s ~3.9s ~4.3s | |
#211 | Command R | 35B | INT4 Q8 | 4 tok/s 3 tok/s | ~31.7s ~31.9s | |
#212 | Llama 3 8B | 8B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#213 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 79 tok/s 51 tok/s 28 tok/s | ~1.7s ~1.7s ~1.8s | |
#214 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 185 tok/s 155 tok/s 112 tok/s | ~273 ms ~274 ms ~276 ms | |
#216 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 162 tok/s 116 tok/s 71 tok/s | ~610 ms ~612 ms ~618 ms | |
#217 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 42 tok/s 25 tok/s 13 tok/s | ~3.7s ~3.7s ~3.7s | |
#218 | Qwen2.5-7B | 7B | INT4 Q8 FP16 | 25 tok/s 14 tok/s 7 tok/s | ~6.5s ~6.6s ~6.6s | |
#219 | Mixtral-8x7B-v0.1 | 46.7B (7B active) | INT4 Q8 | 22 tok/s 13 tok/s | ~7.6s ~7.6s | |
#220 | Ministral-8B-2410 | 8B | INT4 Q8 FP16 | 22 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
#221 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 87 tok/s 56 tok/s 32 tok/s | ~1.5s ~1.5s ~1.6s | |
#222 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
#223 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 42 tok/s 25 tok/s 13 tok/s | ~3.7s ~3.7s ~3.7s | |
#224 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 FP16 | 31 tok/s 25 tok/s 17 tok/s | ~2.2s ~2.2s ~2.2s | |
#225 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 70 tok/s 44 tok/s 24 tok/s | ~2.0s ~2.0s ~2.1s | |
#226 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~6.8s ~6.8s ~6.9s | |
#227 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 115 tok/s 78 tok/s 45 tok/s | ~1.1s ~1.1s ~1.1s | |
#228 | Phi-3-medium | 14B | INT4 Q8 FP16 | 13 tok/s 7 tok/s 4 tok/s | ~12.9s ~13.0s ~13.1s | |
#229 | Yi-34B | 34B | INT4 Q8 FP16 | 4 tok/s 3 tok/s 1 tok/s | ~30.8s ~31.0s ~36.8s | |
#230 | OLMo 3 7B Instruct | 7B | INT4 Q8 FP16 | 17 tok/s 12 tok/s 7 tok/s | ~6.6s ~6.6s ~6.7s | |
#231 | Qwen2-7B | 7B | INT4 Q8 FP16 | 25 tok/s 14 tok/s 7 tok/s | ~6.5s ~6.6s ~6.6s | |
#232 | Phi-3-small | 7B | INT4 Q8 FP16 | 25 tok/s 14 tok/s 7 tok/s | ~6.5s ~6.6s ~6.6s | |
#233 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~6.8s ~6.8s ~6.9s | |
#234 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 115 tok/s 78 tok/s 45 tok/s | ~1.1s ~1.1s ~1.1s | |
#236 | Gemma 1 7B | 7B | INT4 Q8 FP16 | 17 tok/s 12 tok/s 7 tok/s | ~6.6s ~6.6s ~6.7s | |
#237 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 76 tok/s 47 tok/s 25 tok/s | ~2.0s ~2.0s ~2.1s | |
#238 | ChatGLM3-6B | 6B | INT4 Q8 FP16 | 19 tok/s 13 tok/s 8 tok/s | ~5.7s ~5.7s ~5.7s | |
#239 | ChatGLM2-6B | 6B | INT4 Q8 FP16 | 19 tok/s 13 tok/s 8 tok/s | ~5.7s ~5.7s ~5.7s | |
#240 | ChatGLM-6B | 6B | INT4 Q8 FP16 | 19 tok/s 13 tok/s 8 tok/s | ~5.7s ~5.7s ~5.7s | |
未上榜模型（按推理速度排序） | ||||||
- | DINOv3 ViT-L/16 | 0.3B | INT4 Q8 FP16 | 227 tok/s 177 tok/s 119 tok/s | ~302 ms ~304 ms ~306 ms | |
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 180 tok/s 131 tok/s 82 tok/s | ~505 ms ~507 ms ~512 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 180 tok/s 131 tok/s 82 tok/s | ~505 ms ~507 ms ~512 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 173 tok/s 144 tok/s 103 tok/s | ~304 ms ~305 ms ~308 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 173 tok/s 144 tok/s 103 tok/s | ~304 ms ~305 ms ~308 ms | |
- | Liquid D1-Omni 600M | 0.6B | INT4 Q8 FP16 | 162 tok/s 116 tok/s 71 tok/s | ~610 ms ~612 ms ~618 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 126 tok/s 83 tok/s 47 tok/s | ~1.0s ~1.1s ~1.1s | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 115 tok/s 78 tok/s 45 tok/s | ~1.1s ~1.1s ~1.1s | |
- | MiniCPM-V 4.7 1B | 1B | INT4 Q8 FP16 | 115 tok/s 78 tok/s 45 tok/s | ~1.1s ~1.1s ~1.1s | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 87 tok/s 56 tok/s 32 tok/s | ~1.5s ~1.5s ~1.6s | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 87 tok/s 56 tok/s 32 tok/s | ~1.5s ~1.5s ~1.6s | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 79 tok/s 51 tok/s 28 tok/s | ~1.7s ~1.7s ~1.8s | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 76 tok/s 48 tok/s 27 tok/s | ~1.8s ~1.8s ~1.9s | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 55 tok/s 33 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 54 tok/s 43 tok/s 28 tok/s | ~1.3s ~1.4s ~1.4s | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 54 tok/s 43 tok/s 28 tok/s | ~1.3s ~1.4s ~1.4s | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 54 tok/s 43 tok/s 28 tok/s | ~1.3s ~1.4s ~1.4s | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 52 tok/s 32 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | Mellum 2.1 12B A2.5B Thinking | 12B (2.5B active) | INT4 Q8 FP16 | 52 tok/s 34 tok/s 19 tok/s | ~2.9s ~2.9s ~2.9s | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | LFM2-VL 3B | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | Liquid D1 3B | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | Liquid D1-VL 3B | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | Liquid LFM2-VL 3B D1 | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 51 tok/s 31 tok/s 17 tok/s | ~2.9s ~2.9s ~3.0s | |
- | LFM2.5 24B A2B | 24B (2.5B active) | INT4 Q8 FP16 | 47 tok/s 32 tok/s 18 tok/s | ~3.2s ~3.3s ~3.3s | |
- | ERNIE-4.5-21B-A3B | 21B (3B active) | INT4 Q8 FP16 | 43 tok/s 28 tok/s 16 tok/s | ~3.6s ~3.6s ~3.6s | |
- | ERNIE-4.5-21B-A3B-Base | 21B (3B active) | INT4 Q8 FP16 | 43 tok/s 28 tok/s 16 tok/s | ~3.6s ~3.6s ~3.6s | |
- | ERNIE-4.5-VL-28B-A3B | 28B (3B active) | INT4 Q8 FP16 | 41 tok/s 27 tok/s 15 tok/s | ~3.8s ~3.8s ~4.1s | |
- | ERNIE-4.5-VL-28B-A3B-Base | 28B (3B active) | INT4 Q8 FP16 | 41 tok/s 27 tok/s 15 tok/s | ~3.8s ~3.8s ~4.1s | |
- | FrogNano 4B | 4B | INT4 Q8 FP16 | 40 tok/s 24 tok/s 13 tok/s | ~3.8s ~3.9s ~3.9s | |
- | Hy-Memory 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 15 tok/s | ~3.8s ~3.8s ~4.2s | |
- | Hy-MT2 30B A3B | 30B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 15 tok/s | ~3.8s ~3.8s ~4.2s | |
- | MiniCPM-V 4.7 35B A3B | 35B (3B active) | INT4 Q8 FP16 | 40 tok/s 27 tok/s 14 tok/s | ~3.9s ~3.9s ~4.6s | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 40 tok/s 24 tok/s 13 tok/s | ~3.8s ~3.9s ~3.9s | |
- | EmbeddingGemma 2 | 2B | INT4 Q8 FP16 | 38 tok/s 30 tok/s 19 tok/s | ~2.0s ~2.1s ~2.1s | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 33 tok/s 25 tok/s 15 tok/s | ~2.7s ~2.7s ~2.7s | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 31 tok/s 23 tok/s 14 tok/s | ~2.9s ~3.0s ~3.0s | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 31 tok/s 23 tok/s 14 tok/s | ~2.9s ~3.0s ~3.0s | |
- | AliceAI Foundation 80B A3B Base | 80B (3B active) | INT4 | 30 tok/s | ~5.3s | |
- | Kolibri 1 | 78B (3.5B active) | INT4 | 28 tok/s | ~5.7s | |
- | Falcon-7B | 7B | INT4 Q8 FP16 | 26 tok/s 15 tok/s 8 tok/s | ~6.5s ~6.6s ~6.6s | |
- | Falcon3-7B | 7B | INT4 Q8 FP16 | 25 tok/s 14 tok/s 7 tok/s | ~6.5s ~6.6s ~6.6s | |
- | Hy-MT2 7B | 7B | INT4 Q8 FP16 | 25 tok/s 14 tok/s 7 tok/s | ~6.5s ~6.6s ~6.6s | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 FP16 | 24 tok/s 14 tok/s 7 tok/s | ~6.8s ~6.8s ~6.9s | |
- | Youtu Parsing Omni | 8B | INT4 Q8 FP16 | 23 tok/s 13 tok/s 7 tok/s | ~7.4s ~7.5s ~7.6s | |
- | LensVLM 9B | 9B | INT4 Q8 FP16 | 20 tok/s 11 tok/s 6 tok/s | ~8.4s ~8.4s ~8.5s | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 FP16 | 20 tok/s 11 tok/s 6 tok/s | ~8.4s ~8.4s ~8.5s | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 FP16 | 19 tok/s 13 tok/s 8 tok/s | ~5.7s ~5.7s ~5.7s | |
- | Yi-6B | 6B | INT4 Q8 FP16 | 19 tok/s 13 tok/s 8 tok/s | ~5.7s ~5.7s ~5.7s | |
- | Falcon3-10B | 10B | INT4 Q8 FP16 | 18 tok/s 10 tok/s 5 tok/s | ~9.3s ~9.3s ~9.4s | |
- | Kimi-VL-A3B-Instruct | 16B (3B active) | INT4 Q8 FP16 | 18 tok/s 15 tok/s 11 tok/s | ~3.5s ~3.5s ~3.5s | |
- | Falcon2-11B | 11B | INT4 Q8 FP16 | 17 tok/s 10 tok/s 5 tok/s | ~10.2s ~10.2s ~10.3s | |
- | Hunyuan Lite | 7B | INT4 Q8 FP16 | 17 tok/s 12 tok/s 7 tok/s | ~6.6s ~6.6s ~6.7s | |
- | OLMo 3 7B Base | 7B | INT4 Q8 FP16 | 17 tok/s 12 tok/s 7 tok/s | ~6.6s ~6.6s ~6.7s | |
- | SEA-LION-7B | 7.1B | INT4 Q8 FP16 | 17 tok/s 11 tok/s 7 tok/s | ~6.7s ~6.7s ~6.7s | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 FP16 | 17 tok/s 11 tok/s 7 tok/s | ~6.7s ~6.7s ~6.7s | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 FP16 | 15 tok/s 10 tok/s 6 tok/s | ~7.5s ~7.5s ~7.6s | |
- | Typhoon-2-8B | 8B | INT4 Q8 FP16 | 15 tok/s 10 tok/s 6 tok/s | ~7.5s ~7.5s ~7.6s | |
- | GLM-4-9B | 9B | INT4 Q8 FP16 | 14 tok/s 9 tok/s 5 tok/s | ~8.4s ~8.4s ~8.5s | |
- | GLM-4-9B-Chat | 9B | INT4 Q8 FP16 | 14 tok/s 9 tok/s 5 tok/s | ~8.4s ~8.4s ~8.5s | |
- | GLM-4-9B-Chat-1M | 9B | INT4 Q8 FP16 | 14 tok/s 9 tok/s 5 tok/s | ~8.4s ~8.4s ~8.5s | |
- | Laguna S 2.1 | 118B (8B active) | INT4 | 14 tok/s | ~12.0s | |
- | Yi-9B | 9B | INT4 Q8 FP16 | 14 tok/s 9 tok/s 5 tok/s | ~8.4s ~8.4s ~8.5s | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~8.6s ~8.6s ~8.7s | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 Q8 FP16 | 13 tok/s 9 tok/s 5 tok/s | ~8.6s ~8.6s ~8.7s | |
- | GreenMind-14B-R1 | 14B | INT4 Q8 FP16 | 9 tok/s 6 tok/s 3 tok/s | ~12.9s ~13.0s ~13.1s | |
- | AREX-2 | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
- | PPLX Decider v1 27B | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
- | PPLX Decider v1.1 27B | 27B | INT4 Q8 FP16 | 7 tok/s 4 tok/s 2 tok/s | ~24.6s ~24.7s ~24.9s | |
- | Agnes 3.0 Flash | 33B | INT4 Q8 FP16 | 6 tok/s 3 tok/s 2 tok/s | ~29.9s ~30.0s ~33.0s | |
- | Falcon-40B | 40B | INT4 Q8 | 5 tok/s 3 tok/s | ~36.1s ~36.3s | |
- | Hunyuan A13B | 80B (13B active) | INT4 | 5 tok/s | ~14.0s | |
- | Kimi-Dev-72B | 72B | INT4 | 2 tok/s | ~65.0s | |
- | Typhoon-2-70B | 70B | INT4 | 2 tok/s | ~63.1s | |
推理能力评估
不同量化精度下可运行的最大模型参数级别。
上下文长度:
4位量化
70B+ 模型
8位精度
32B 模型
16位全精度
32B 模型
微调能力评估
本地模型训练与适配器微调支持能力。
微调上下文:
QLoRA
70B+ 模型
LoRA
70B 模型
全参数微调
7B-8B 模型
负载与选型建议
AMD Ryzen AI 9 HX 370 (96GB system, 72GB GPU-accessible) 上的最佳量化精度与运行规格建议。
最佳推理区间
适合在 Q4/Q8 精度下运行 70B 参数级模型并支持 32k 上下文扩展，或对中小型模型进行高并发批处理推理。
显存带宽与推理速率
具备 128 GB/s 聚合带宽，单批次推理处于显存带宽受限模式。8B Q4 模型预估生成速度约 28 tok/s，70B Q4 模型预估生成速度约 3 tok/s。