趋近智
AMD RX 6750 XT (12GB)
AMD RX 6750 XT (12GB) 提供 12 GB 显存/内存与 432 GB/s 带宽，支持通过 ROCm 与 Vulkan 加速进行本地推理。
总显存
12 GB
可用显存上限
12 GB
显存带宽
432 GB/s
最大推理模型级别
14B 模型 (Q4)
热设计功耗
250W
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 RX 6750 XT (12GB) 最佳运行大模型与速度矩阵
主流开源大模型在 AMD RX 6750 XT (12GB) 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 |
|---|---|---|---|---|---|---|
#112 | Phi-4 Reasoning Plus | 14B | INT4 | 26 tok/s | ~7.6s | |
#125 | Qwen3.5-9B | 9B | INT4 Q8 | 56 tok/s 33 tok/s | ~4.2s ~4.6s | |
#128 | K2 Horizon 7B | 7B | INT4 Q8 | 68 tok/s 43 tok/s | ~3.3s ~3.3s | |
#135 | Gemma 4 12B | 11.95B | INT4 | 31 tok/s | ~6.0s | |
#140 | Qwen3.5-4B | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
#144 | GLM-4V | 9B | INT4 | 41 tok/s | ~4.2s | |
#150 | NVIDIA Nemotron 3 Nano 30B-A3B | 3.5B (30B active) | INT4 Q8 FP16 | 19 tok/s 11 tok/s 6 tok/s | ~13.2s ~13.3s ~14.5s | |
#158 | Granite 4.2 8B | 8B | INT4 Q8 | 62 tok/s 36 tok/s | ~3.7s ~4.1s | |
#159 | Gemma 4 E4B | 8B | INT4 Q8 | 62 tok/s 36 tok/s | ~3.7s ~4.1s | |
#166 | Phi-4 | 14B | INT4 | 39 tok/s | ~6.5s | |
#178 | OLMo 3 7B Think | 7B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
#179 | DeepSeek-R1 14B | 14B | INT4 | 40 tok/s | ~6.5s | |
#183 | Ministral 3 14B | 14B | INT4 | 26 tok/s | ~7.6s | |
#186 | Gemma 3 12B | 12B | INT4 | 45 tok/s | ~5.6s | |
#189 | Ministral 3 8B | 8B | INT4 Q8 | 45 tok/s 29 tok/s | ~3.8s ~4.4s | |
#191 | MiniCPM5-2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
#192 | Qwen3.5-2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
#194 | Qwen2.5-14B | 14B | INT4 | 39 tok/s | ~6.5s | |
#200 | Gemma 4 E2B | 5.1B | INT4 Q8 FP16 | 85 tok/s 56 tok/s 28 tok/s | ~2.4s ~2.4s ~2.9s | |
#202 | DeepSeek-R1 8B | 8B | INT4 Q8 | 63 tok/s 39 tok/s | ~3.7s ~3.8s | |
#204 | Qwen3-4B | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
#205 | Ministral 3 3B | 3B | INT4 Q8 FP16 | 81 tok/s 64 tok/s 42 tok/s | ~1.5s ~1.5s ~1.5s | |
#207 | Gemma 2 9B | 9B | INT4 Q8 | 56 tok/s 33 tok/s | ~4.2s ~4.6s | |
#208 | Qwen3.5-0.8B | 0.8B | INT4 Q8 FP16 | 204 tok/s 170 tok/s 124 tok/s | ~420 ms ~421 ms ~423 ms | |
#209 | Qwen3-14B | 14B | INT4 | 39 tok/s | ~6.5s | |
#210 | Gemma 3 4B | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
#211 | Qwen3-8B | 8B | INT4 Q8 | 62 tok/s 36 tok/s | ~3.7s ~4.1s | |
#213 | DeepSeek-R1 7B | 7B | INT4 Q8 | 70 tok/s 44 tok/s | ~3.3s ~3.3s | |
#214 | Qwen3-1.7B | 1.7B | INT4 Q8 FP16 | 155 tok/s 117 tok/s 76 tok/s | ~878 ms ~880 ms ~885 ms | |
#215 | Llama 3.1 8B | 8B | INT4 Q8 | 62 tok/s 36 tok/s | ~3.7s ~4.1s | |
#216 | Gemma 3 270M | 0.27B | INT4 Q8 FP16 | 232 tok/s 217 tok/s 187 tok/s | ~142 ms ~143 ms ~143 ms | |
#218 | Qwen3-0.6B | 0.6B | INT4 Q8 FP16 | 221 tok/s 190 tok/s 145 tok/s | ~312 ms ~312 ms ~314 ms | |
#219 | Llama 3 8B | 8B | INT4 Q8 | 62 tok/s 36 tok/s | ~3.7s ~4.1s | |
#221 | Phi-4-Mini | 3.8B | INT4 Q8 FP16 | 103 tok/s 69 tok/s 40 tok/s | ~1.8s ~1.8s ~1.9s | |
#223 | Qwen2.5-7B | 7B | INT4 Q8 | 68 tok/s 43 tok/s | ~3.3s ~3.3s | |
#224 | Ministral-8B-2410 | 8B | INT4 Q8 | 62 tok/s 36 tok/s | ~3.7s ~4.1s | |
#225 | DeepSeek-R1 1.5B | 1.5B | INT4 Q8 FP16 | 164 tok/s 126 tok/s 83 tok/s | ~778 ms ~779 ms ~783 ms | |
#226 | Llama 3.2 3B | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
#228 | Phi-3-mini | 3.8B | INT4 Q8 FP16 | 103 tok/s 69 tok/s 40 tok/s | ~1.8s ~1.8s ~1.9s | |
#229 | Gemma 3n E2B IT | 6B (2B active) | INT4 Q8 | 81 tok/s 70 tok/s | ~1.1s ~1.1s | |
#230 | Mistral-7B-Instruct-v0.1 | 7.3B | INT4 Q8 | 66 tok/s 41 tok/s | ~3.4s ~3.4s | |
#231 | Llama 3.2 1B | 1B | INT4 Q8 FP16 | 189 tok/s 154 tok/s 108 tok/s | ~533 ms ~534 ms ~536 ms | |
#232 | Gemma 2 2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
#233 | Phi-3-medium | 14B | INT4 | 39 tok/s | ~6.5s | |
#234 | OLMo 3 7B Instruct | 7B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
#236 | Qwen2-7B | 7B | INT4 Q8 | 68 tok/s 43 tok/s | ~3.3s ~3.3s | |
#237 | Phi-3-small | 7B | INT4 Q8 | 68 tok/s 43 tok/s | ~3.3s ~3.3s | |
#238 | Gemma 3 1B | 1B | INT4 Q8 FP16 | 189 tok/s 154 tok/s 108 tok/s | ~533 ms ~534 ms ~536 ms | |
#239 | Mistral-7B-Instruct-v0.2 | 7.3B | INT4 Q8 | 66 tok/s 41 tok/s | ~3.4s ~3.4s | |
#241 | Gemma 1 7B | 7B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
#242 | Gemma 1 2B | 2B | INT4 Q8 FP16 | 152 tok/s 110 tok/s 69 tok/s | ~1.0s ~1.0s ~1.0s | |
#243 | ChatGLM3-6B | 6B | INT4 Q8 | 55 tok/s 40 tok/s | ~2.8s ~2.8s | |
#244 | ChatGLM2-6B | 6B | INT4 Q8 | 55 tok/s 40 tok/s | ~2.8s ~2.8s | |
#245 | ChatGLM-6B | 6B | INT4 Q8 | 55 tok/s 40 tok/s | ~2.8s ~2.8s | |
未上榜模型（按推理速度排序） | ||||||
- | DINOv3 ViT-L/16 | 0.3B | INT4 Q8 FP16 | 250 tok/s 229 tok/s 193 tok/s | ~157 ms ~158 ms ~159 ms | |
- | Qwen2-0.5B | 0.5B | INT4 Q8 FP16 | 230 tok/s 202 tok/s 158 tok/s | ~259 ms ~260 ms ~261 ms | |
- | Qwen2.5-0.5B | 0.5B | INT4 Q8 FP16 | 230 tok/s 202 tok/s 158 tok/s | ~259 ms ~260 ms ~261 ms | |
- | ERNIE-4.5-0.3B | 0.3B | INT4 Q8 FP16 | 226 tok/s 210 tok/s 180 tok/s | ~158 ms ~158 ms ~159 ms | |
- | ERNIE-4.5-0.3B-Base | 0.3B | INT4 Q8 FP16 | 226 tok/s 210 tok/s 180 tok/s | ~158 ms ~158 ms ~159 ms | |
- | Liquid D1-Omni 600M | 0.6B | INT4 Q8 FP16 | 221 tok/s 190 tok/s 145 tok/s | ~312 ms ~312 ms ~314 ms | |
- | PixArt-α XL 2 1024-MS | 0.61B | INT4 Q8 FP16 | 220 tok/s 189 tok/s 144 tok/s | ~316 ms ~317 ms ~319 ms | |
- | PixArt-Σ XL 2 1024-MS | 0.61B | INT4 Q8 FP16 | 220 tok/s 189 tok/s 144 tok/s | ~316 ms ~317 ms ~319 ms | |
- | Nemotron 3.5 Parse | 0.63B | INT4 Q8 FP16 | 218 tok/s 187 tok/s 141 tok/s | ~329 ms ~329 ms ~331 ms | |
- | SD 2.1 Turbo | 0.86B | INT4 Q8 FP16 | 200 tok/s 165 tok/s 119 tok/s | ~453 ms ~454 ms ~456 ms | |
- | Stable Diffusion 1.4 | 0.86B | INT4 Q8 FP16 | 200 tok/s 165 tok/s 119 tok/s | ~453 ms ~454 ms ~456 ms | |
- | Stable Diffusion 2.1 | 0.86B | INT4 Q8 FP16 | 200 tok/s 165 tok/s 119 tok/s | ~453 ms ~454 ms ~456 ms | |
- | Falcon-1B | 1B | INT4 Q8 FP16 | 198 tok/s 159 tok/s 111 tok/s | ~532 ms ~534 ms ~536 ms | |
- | Falcon3-1B | 1B | INT4 Q8 FP16 | 189 tok/s 154 tok/s 108 tok/s | ~533 ms ~534 ms ~536 ms | |
- | MiniCPM-V 4.7 1B | 1B | INT4 Q8 FP16 | 189 tok/s 154 tok/s 108 tok/s | ~533 ms ~534 ms ~536 ms | |
- | Kandinsky 2.2 Decoder | 1.22B | INT4 Q8 FP16 | 177 tok/s 140 tok/s 96 tok/s | ~641 ms ~642 ms ~646 ms | |
- | Wan 2.1 T2V 1.3B | 1.3B | INT4 Q8 FP16 | 173 tok/s 136 tok/s 92 tok/s | ~678 ms ~679 ms ~683 ms | |
- | Qwen2-1.5B | 1.5B | INT4 Q8 FP16 | 164 tok/s 126 tok/s 83 tok/s | ~778 ms ~779 ms ~783 ms | |
- | Qwen2.5-1.5B | 1.5B | INT4 Q8 FP16 | 164 tok/s 126 tok/s 83 tok/s | ~778 ms ~779 ms ~783 ms | |
- | SVD Image-to-Video XT | 1.5B | INT4 Q8 FP16 | 164 tok/s 126 tok/s 83 tok/s | ~778 ms ~779 ms ~783 ms | |
- | SVD Image-to-Video XT 1.1 | 1.5B | INT4 Q8 FP16 | 164 tok/s 126 tok/s 83 tok/s | ~778 ms ~779 ms ~783 ms | |
- | LFM3 1.7B | 1.7B | INT4 Q8 FP16 | 155 tok/s 117 tok/s 76 tok/s | ~878 ms ~880 ms ~885 ms | |
- | LFM3 1.7B Base | 1.7B | INT4 Q8 FP16 | 155 tok/s 117 tok/s 76 tok/s | ~878 ms ~880 ms ~885 ms | |
- | Spark X2.5 1.7B | 1.7B | INT4 Q8 FP16 | 155 tok/s 117 tok/s 76 tok/s | ~878 ms ~880 ms ~885 ms | |
- | Hy-MT2 1.8B | 1.8B | INT4 Q8 FP16 | 151 tok/s 113 tok/s 73 tok/s | ~925 ms ~927 ms ~932 ms | |
- | CogVideoX 2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | LFM3 Audio 2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | LFM3 Omni 2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | LFM3 VL 2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | Liquid D1 2B | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | LTX-Video 0.9.1 | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | Moondream 3.1 | 9B (2B active) | INT4 Q8 | 144 tok/s 101 tok/s | ~1.2s ~1.3s | |
- | Stable Diffusion 3 Medium | 2B | INT4 Q8 FP16 | 144 tok/s 106 tok/s 67 tok/s | ~1.0s ~1.0s ~1.0s | |
- | Kandinsky 3.0 | 2.1B | INT4 Q8 FP16 | 141 tok/s 103 tok/s 65 tok/s | ~1.1s ~1.1s ~1.1s | |
- | LTX-2 | 2.2B | INT4 Q8 FP16 | 138 tok/s 100 tok/s 63 tok/s | ~1.1s ~1.1s ~1.1s | |
- | SDXL 1.0 Refiner | 2.31B | INT4 Q8 FP16 | 135 tok/s 97 tok/s 60 tok/s | ~1.2s ~1.2s ~1.2s | |
- | Stable Diffusion 3.5 Medium | 2.5B | INT4 Q8 FP16 | 130 tok/s 93 tok/s 57 tok/s | ~1.3s ~1.3s ~1.3s | |
- | SDXL 1.0 Base | 2.56B | INT4 Q8 FP16 | 128 tok/s 91 tok/s 56 tok/s | ~1.3s ~1.3s ~1.3s | |
- | SDXL Turbo | 2.56B | INT4 Q8 FP16 | 128 tok/s 91 tok/s 56 tok/s | ~1.3s ~1.3s ~1.3s | |
- | Falcon-3B | 3B | INT4 Q8 FP16 | 124 tok/s 85 tok/s 50 tok/s | ~1.5s ~1.5s ~1.5s | |
- | CroissantLLM Base | 1.3B | INT4 Q8 FP16 | 122 tok/s 104 tok/s 76 tok/s | ~680 ms ~681 ms ~685 ms | |
- | Phi-1 | 1.3B | INT4 Q8 FP16 | 122 tok/s 104 tok/s 76 tok/s | ~680 ms ~681 ms ~685 ms | |
- | Phi-1.5 | 1.3B | INT4 Q8 FP16 | 122 tok/s 104 tok/s 76 tok/s | ~680 ms ~681 ms ~685 ms | |
- | DeepSeek-R1 3B | 3B | INT4 Q8 FP16 | 119 tok/s 83 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | Mellum 2.1 12B A2.5B Thinking | 12B (2.5B active) | INT4 | 119 tok/s | ~1.4s | |
- | Falcon3-3B | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | LFM2-VL 3B | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | Liquid D1 3B | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | Liquid D1-VL 3B | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | Liquid LFM2-VL 3B D1 | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | Qwen2.5-3B | 3B | INT4 Q8 FP16 | 118 tok/s 82 tok/s 49 tok/s | ~1.5s ~1.5s ~1.5s | |
- | FLUX.2 Klein 4B | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
- | FrogNano 4B | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Mage-VL Cowork | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
- | Spark X2.5 4B | 4B | INT4 Q8 FP16 | 100 tok/s 67 tok/s 39 tok/s | ~1.9s ~1.9s ~1.9s | |
- | EmbeddingGemma 2 | 2B | INT4 Q8 FP16 | 96 tok/s 79 tok/s 55 tok/s | ~1.0s ~1.0s ~1.0s | |
- | CogVideoX 1.5 5B | 5B | INT4 Q8 FP16 | 86 tok/s 56 tok/s 30 tok/s | ~2.4s ~2.4s ~2.6s | |
- | CogVideoX 5B | 5B | INT4 Q8 FP16 | 86 tok/s 56 tok/s 30 tok/s | ~2.4s ~2.4s ~2.6s | |
- | Wan 2.2 TI2V 5B | 5B | INT4 Q8 FP16 | 86 tok/s 56 tok/s 30 tok/s | ~2.4s ~2.4s ~2.6s | |
- | Phi-2 | 2.7B | INT4 Q8 FP16 | 85 tok/s 68 tok/s 45 tok/s | ~1.3s ~1.3s ~1.4s | |
- | MaLLaM-3B | 3B | INT4 Q8 FP16 | 81 tok/s 64 tok/s 42 tok/s | ~1.5s ~1.5s ~1.5s | |
- | SmolLM3 3B | 3B | INT4 Q8 FP16 | 81 tok/s 64 tok/s 42 tok/s | ~1.5s ~1.5s ~1.5s | |
- | Falcon-7B | 7B | INT4 Q8 | 71 tok/s 44 tok/s | ~3.3s ~3.3s | |
- | AuraFlow v0.2 | 6.8B | INT4 Q8 | 70 tok/s 44 tok/s | ~3.2s ~3.2s | |
- | AuraFlow v0.3 | 6.8B | INT4 Q8 | 70 tok/s 44 tok/s | ~3.2s ~3.2s | |
- | Falcon3-7B | 7B | INT4 Q8 | 68 tok/s 43 tok/s | ~3.3s ~3.3s | |
- | Hy-MT2 7B | 7B | INT4 Q8 | 68 tok/s 43 tok/s | ~3.3s ~3.3s | |
- | Mistral-7B-v0.1 | 7.3B | INT4 Q8 | 66 tok/s 41 tok/s | ~3.4s ~3.4s | |
- | Youtu Parsing Omni | 8B | INT4 Q8 | 63 tok/s 39 tok/s | ~3.7s ~3.8s | |
- | Stable Diffusion 3.5 Large | 8.1B | INT4 Q8 | 61 tok/s 36 tok/s | ~3.8s ~4.1s | |
- | Stable Diffusion 3.5 Large Turbo | 8.1B | INT4 Q8 | 61 tok/s 36 tok/s | ~3.8s ~4.1s | |
- | FLUX.2 Klein 9B | 9B | INT4 Q8 | 56 tok/s 33 tok/s | ~4.2s ~4.6s | |
- | LensVLM 9B | 9B | INT4 Q8 | 56 tok/s 33 tok/s | ~4.2s ~4.6s | |
- | MiMo V2.6 Distill Qwen 9B | 9B | INT4 Q8 | 56 tok/s 33 tok/s | ~4.2s ~4.6s | |
- | ChatGLM3-6B-32K | 6B | INT4 Q8 | 55 tok/s 40 tok/s | ~2.8s ~2.8s | |
- | Yi-6B | 6B | INT4 Q8 | 55 tok/s 40 tok/s | ~2.8s ~2.8s | |
- | Falcon3-10B | 10B | INT4 Q8 | 52 tok/s 28 tok/s | ~4.7s ~5.5s | |
- | Mochi 1 Preview | 10B | INT4 Q8 | 52 tok/s 28 tok/s | ~4.7s ~5.5s | |
- | Falcon2-11B | 11B | INT4 Q8 | 50 tok/s 27 tok/s | ~5.1s ~6.0s | |
- | Hunyuan Lite | 7B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
- | OLMo 3 7B Base | 7B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
- | SEA-LION-7B | 7.1B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
- | SEA-LION-7B-Instruct | 7.1B | INT4 Q8 | 49 tok/s 33 tok/s | ~3.3s ~3.6s | |
- | FLUX.1 Canny Dev | 11.9B | INT4 | 45 tok/s | ~5.5s | |
- | FLUX.1 Depth Dev | 11.9B | INT4 | 45 tok/s | ~5.5s | |
- | FLUX.1 Dev | 11.9B | INT4 | 45 tok/s | ~5.5s | |
- | FLUX.1 Fill Dev | 11.9B | INT4 | 45 tok/s | ~5.5s | |
- | FLUX.1 Schnell | 11.9B | INT4 | 45 tok/s | ~5.5s | |
- | Sahabat-AI-Llama3-8B-Instruct | 8B | INT4 Q8 | 45 tok/s 29 tok/s | ~3.8s ~4.4s | |
- | Typhoon-2-8B | 8B | INT4 Q8 | 45 tok/s 29 tok/s | ~3.8s ~4.4s | |
- | HunyuanVideo 1.5 720P T2V | 13B | INT4 | 42 tok/s | ~6.0s | |
- | GLM-4-9B | 9B | INT4 | 41 tok/s | ~4.2s | |
- | GLM-4-9B-Chat | 9B | INT4 | 41 tok/s | ~4.2s | |
- | GLM-4-9B-Chat-1M | 9B | INT4 | 41 tok/s | ~4.2s | |
- | Yi-9B | 9B | INT4 | 41 tok/s | ~4.2s | |
- | Sahabat-AI-Gemma2-9B | 9.2B | INT4 | 40 tok/s | ~4.3s | |
- | Sahabat-AI-Gemma2-9B-Instruct | 9.2B | INT4 | 40 tok/s | ~4.3s | |
- | Wan 2.1 I2V 14B 720P | 14B | INT4 | 39 tok/s | ~6.5s | |
- | Wan 2.1 T2V 14B | 14B | INT4 | 39 tok/s | ~6.5s | |
- | Wan 2.2 T2V A14B | 14B | INT4 | 39 tok/s | ~6.5s | |
- | FLUX.2 Dev | 15B | INT4 | 37 tok/s | ~6.9s | |
- | GreenMind-14B-R1 | 14B | INT4 | 26 tok/s | ~7.6s | |
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
14B 模型
8位精度
8B 模型
16位全精度
3B 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
14B 模型
LoRA
8B 模型
全参数微调
1B 模型
负载与选型建议
AMD RX 6750 XT (12GB) 上的最佳量化精度与运行规格建议。
最佳推理区间
非常适合 8B 模型长上下文推理（32k+）。14B 模型在 Q4 精度及标准上下文下也能流畅运行。
显存带宽与推理速率
具备 432 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 96 tok/s，70B Q4 模型预估生成速度约 11 tok/s。