{"id":82094,"date":"2026-07-25T06:36:45","date_gmt":"2026-07-24T22:36:45","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/82094.html"},"modified":"2026-07-25T06:36:45","modified_gmt":"2026-07-24T22:36:45","slug":"amd-ryzen-ai-strix-halo%e6%9e%b6%e6%9e%84%e5%a4%84%e7%90%86%e5%99%a8%ef%bc%9a%e5%a6%82%e4%bd%95%e5%9c%a8%e7%ac%94%e8%ae%b0%e6%9c%ac%e4%b8%8a%e8%b7%91%e9%80%9a%e5%8e%9f%e6%9c%ac%e5%b1%9e%e4%ba%8e","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/82094.html","title":{"rendered":"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f"},"content":{"rendered":"<p>\u672c\u6587\u5c06\u7cfb\u7edf\u5730\u4ecb\u7ecd\u4e00\u4e0b\u5982\u4f55\u5728 Strix Halo \u5e73\u53f0\u4e0a\u90e8\u7f72\u670d\u52a1\u5668\u7ea7 AI \u5927\u6a21\u578b\u3002\u5148\u753b\u4e00\u5f20\u6574\u4f53\u67b6\u6784\u56fe&#xff0c;\u518d\u8be6\u7ec6\u5c55\u5f00\u3002<br \/>\n<img decoding=\"async\" src=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/07\/20260724223643-6a63e8fb7f4c9.png\" alt=\"\u5728\u8fd9\u91cc\u63d2\u5165\u56fe\u7247\u63cf\u8ff0\" \/><\/p>\n<h3>\u4e00\u3001\u4e3a\u4ec0\u4e48 Strix Halo \u80fd\u8dd1\u670d\u52a1\u5668\u7ea7\u5927\u6a21\u578b&#xff1f;<\/h3>\n<h4>\u6838\u5fc3\u4f18\u52bf&#xff1a;\u7edf\u4e00\u5927\u5185\u5b58\u7a81\u7834\u663e\u5b58\u74f6\u9888<\/h4>\n<p>\u4f20\u7edf\u6d88\u8d39\u7ea7 GPU&#xff08;RTX 4090&#xff09;\u6700\u591a\u53ea\u6709 24 GB \u72ec\u7acb\u663e\u5b58&#xff0c;\u65e0\u6cd5\u5728\u5355\u5361\u4e0a\u8fd0\u884c 70B \u53c2\u6570\u6a21\u578b\u3002Strix Halo&#xff08;Ryzen AI Max&#043; 395&#xff09;\u6700\u591a\u652f\u6301 128 GB LPDDR5X-8000 \u7edf\u4e00\u5185\u5b58&#xff0c;CPU \u548c iGPU \u5171\u4eab\u8fd9\u5757\u5185\u5b58&#xff0c;\u56e0\u6b64\u53ef\u4ee5\u5b8c\u6574\u52a0\u8f7d Llama 3.1 70B \u7684 BF16 \u5168\u7cbe\u5ea6\u6743\u91cd&#xff0c;\u800c\u6ca1\u6709\u4efb\u4f55\u6d88\u8d39\u7ea7 NVIDIA \u663e\u5361\u80fd\u505a\u5230\u8fd9\u4e00\u70b9\u3002<\/p>\n<p>\u4e0e\u72ec\u7acb\u663e\u5361\u7684\u4e13\u7528 VRAM \u4e0d\u540c&#xff0c;Ryzen AI Max \u91c7\u7528\u7edf\u4e00\u5185\u5b58\u67b6\u6784&#xff0c;\u901a\u8fc7 GTT&#xff08;\u56fe\u5f62\u8f6c\u6362\u8868&#xff09;\u8ba9 GPU \u8bbf\u95ee\u7cfb\u7edf\u5185\u5b58&#xff0c;\u914d\u7f6e\u540e\u7ea6\u6709 96 GB \u53ef\u7528\u4e8e LLM \u63a8\u7406&#xff0c;\u53ef\u4ee5\u52a0\u8f7d\u539f\u672c\u9700\u8981\u591a\u5757\u9ad8\u7aef GPU \u624d\u80fd\u8fd0\u884c\u7684\u6a21\u578b\u3002<\/p>\n<hr \/>\n<h3>\u4e8c\u3001\u7cfb\u7edf\u51c6\u5907&#xff1a;BIOS \u4e0e\u5185\u6838\u914d\u7f6e&#xff08;Linux \u6700\u4f18&#xff09;<\/h3>\n<p>BIOS \u5173\u952e\u914d\u7f6e&#xff1a;<\/p>\n<li>UMA&#xff08;\u7edf\u4e00\u5185\u5b58\u67b6\u6784&#xff09;\u5206\u914d&#xff1a;\u5c06 GPU \u4e13\u7528\u663e\u5b58\u5206\u533a\u8bbe\u4e3a 96 GB&#xff08;\u6216\u5c3d\u91cf\u5927&#xff09;&#xff0c;\u8fd9\u76f4\u63a5\u51b3\u5b9a GPU \u53ef\u89c1\u7684\u663e\u5b58\u6c60\u5927\u5c0f\u3002<\/li>\n<li>TDP \u529f\u8017\u914d\u7f6e&#xff1a;\u5efa\u8bae\u8bbe\u4e3a 85 W&#xff0c;\u8fd9\u662f\u6027\u80fd\/\u6548\u7387\u7684\u6700\u4f18\u5e73\u8861\u70b9&#xff1b;24 \u5c0f\u65f6\u63a8\u7406\u670d\u52a1\u5668\u53ef\u5728 65\u201390 W \u533a\u95f4\u8fd0\u884c\u3002<\/li>\n<li>IOMMU&#xff1a;\u7981\u7528&#xff0c;\u907f\u514d\u5185\u5b58\u8bbf\u95ee\u88ab\u9650\u5236\u3002<\/li>\n<p>Linux \u5185\u6838\u8981\u6c42&#xff1a;<\/p>\n<p>\u9700\u8981\u5185\u6838 6.16.9 \u6216\u66f4\u9ad8\u7248\u672c\u624d\u80fd\u83b7\u5f97\u5b8c\u6574\u7684\u5185\u5b58\u8bbf\u95ee\u652f\u6301\u3002\u5982\u679c\u7cfb\u7edf\u5185\u6838\u8f83\u65e7&#xff0c;\u53ef\u901a\u8fc7 mainline \u5de5\u5177\u5b89\u88c5\u65b0\u5185\u6838&#xff1a;<\/p>\n<p><span class=\"token function\">sudo<\/span> add-apt-repository ppa:cappelikan\/ppa <span class=\"token parameter variable\">-y<\/span><br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt<\/span> update<br \/>\n<span class=\"token function\">sudo<\/span> <span class=\"token function\">apt<\/span> <span class=\"token function\">install<\/span> mainline <span class=\"token parameter variable\">-y<\/span><br \/>\n<span class=\"token function\">sudo<\/span> mainline <span class=\"token parameter variable\">&#8211;install<\/span> <span class=\"token number\">6.16<\/span>.9<\/p>\n<hr \/>\n<h3>\u4e09\u3001\u9a71\u52a8\u5c42&#xff1a;Vulkan RADV vs ROCm&#xff0c;\u5982\u4f55\u9009\u62e9&#xff1f;<\/h3>\n<p>AMDVLK&#xff08;AMD \u4ee5\u524d\u7684\u5f00\u6e90 Vulkan \u9a71\u52a8&#xff09;\u5df2\u4e8e 2025 \u5e74 4 \u6708\u505c\u6b62\u66f4\u65b0&#xff0c;RADV \u73b0\u5728\u662f AMD \u552f\u4e00\u652f\u6301\u7684\u5f00\u6e90 Vulkan \u9a71\u52a8\u3002RADV \u5728\u63d0\u793a\u8bcd\u5904\u7406&#xff08;pp&#xff09;\u4e0a\u6bd4 AMDVLK \u5feb 63%&#xff0c;\u4e14\u89e3\u7801\u901f\u5ea6\u4e5f\u66f4\u5feb\u3002\u7279\u522b\u6ce8\u610f&#xff1a;\u5373\u4f7f\u4e0d\u4e3b\u52a8\u4f7f\u7528 AMDVLK&#xff0c;\u5176 ICD \u914d\u7f6e\u6587\u4ef6\u4ecd\u4f1a\u9759\u9ed8\u52ab\u6301 Vulkan&#xff0c;\u5c06\u63d0\u793a\u8bcd\u5904\u7406\u901f\u5ea6\u51cf\u534a\u2014\u2014\u5e94\u5f7b\u5e95\u5378\u8f7d\u3002<\/p>\n<p>\u5b9e\u9645\u9009\u62e9\u5efa\u8bae&#xff1a;<\/p>\n<table>\n<tr>\u573a\u666f\u63a8\u8350\u540e\u7aef<\/tr>\n<tbody>\n<tr>\n<td>\u65e5\u5e38 LLM \u63a8\u7406&#xff08;\u77ed\u4e0a\u4e0b\u6587&#xff09;<\/td>\n<td>Vulkan RADV<\/td>\n<\/tr>\n<tr>\n<td>\u9700\u8981 vLLM \u517c\u5bb9<\/td>\n<td>ROCm HIP<\/td>\n<\/tr>\n<tr>\n<td>MoE \u5927\u6a21\u578b<\/td>\n<td>Vulkan &#043; Flash Attention<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>\u4f7f\u7528 ROCm HIP \u65f6\u9700\u8981\u8bbe\u7f6e\u73af\u5883\u53d8\u91cf HSA_OVERRIDE_GFX_VERSION&#061;11.5.1&#xff0c;Ollama 0.20&#043; \u5df2\u4fee\u590d\u6b64\u95ee\u9898\u3002<\/p>\n<hr \/>\n<h3>\u56db\u3001\u63a8\u7406\u6846\u67b6\u8be6\u7ec6\u90e8\u7f72<\/h3>\n<h4>\u65b9\u6848 A&#xff1a;Ollama&#xff08;\u6700\u5feb\u4e0a\u624b&#xff09;<\/h4>\n<p><span class=\"token comment\"># \u5b89\u88c5 Ollama<\/span><br \/>\n<span class=\"token function\">curl<\/span> <span class=\"token parameter variable\">-fsSL<\/span> https:\/\/ollama.com\/install.sh <span class=\"token operator\">|<\/span> <span class=\"token function\">sh<\/span><\/p>\n<p><span class=\"token comment\"># \u542f\u7528 Flash Attention&#xff08;\u5173\u952e\u4f18\u5316&#xff09;<\/span><br \/>\n<span class=\"token builtin class-name\">export<\/span> <span class=\"token assign-left variable\">OLLAMA_FLASH_ATTENTION<\/span><span class=\"token operator\">&#061;<\/span><span class=\"token number\">1<\/span><\/p>\n<p><span class=\"token comment\"># \u76f4\u63a5\u62c9\u53d6\u8fd0\u884c 70B \u6a21\u578b<\/span><br \/>\nollama run llama3.1:70b<br \/>\nollama run qwen3:30b-a3b   <span class=\"token comment\"># MoE \u7ed3\u6784&#xff0c;\u901f\u5ea6\u66f4\u5feb<\/span><\/p>\n<p>\u542f\u7528 Flash Attention \u662f\u5bf9 Strix Halo \u6700\u91cd\u8981\u7684\u5355\u9879\u4f18\u5316&#xff1b;Ollama 0.21 \u7248\u672c\u5f00\u59cb\u9ed8\u8ba4\u542f\u7528\u3002<\/p>\n<h4>\u65b9\u6848 B&#xff1a;llama.cpp &#043; ROCm&#xff08;\u6027\u80fd\u6700\u5f3a&#xff09;<\/h4>\n<p>AMD \u5b98\u65b9\u63d0\u4f9b\u4e86\u9762\u5411 gfx1151&#xff08;Strix Halo GPU \u4ee3\u53f7&#xff09;\u7684\u9884\u7f16\u8bd1\u4e8c\u8fdb\u5236&#xff1a;<\/p>\n<p><span class=\"token comment\"># \u4f7f\u7528 Lemonade SDK \u9884\u7f16\u8bd1\u5305&#xff08;\u514d\u53bb\u81ea\u884c\u7f16\u8bd1&#xff09;<\/span><br \/>\n<span class=\"token function\">wget<\/span> https:\/\/github.com\/lemonade-sdk\/llamacpp-rocm\/releases\/latest\/download\/<span class=\"token punctuation\">\\\\<\/span><br \/>\nllama-bXXXX-ubuntu-rocm-gfx1151-x64.zip<br \/>\n<span class=\"token function\">unzip<\/span> llama-bXXXX-ubuntu-rocm-gfx1151-x64.zip<br \/>\n<span class=\"token function\">chmod<\/span> &#043;x llama-cli llama-server<\/p>\n<p><span class=\"token comment\"># \u9a8c\u8bc1 GPU \u8bc6\u522b<\/span><br \/>\n.\/llama-cli &#8211;list-devices<br \/>\n<span class=\"token comment\"># \u5e94\u8f93\u51fa: Device 0: AMD Radeon Graphics, gfx1151<\/span><\/p>\n<p><span class=\"token comment\"># \u8fd0\u884c 70B \u6a21\u578b&#xff0c;\u5f00\u542f Flash Attention<\/span><br \/>\n.\/llama-server <span class=\"token punctuation\">\\\\<\/span><br \/>\n  <span class=\"token parameter variable\">-m<\/span> Llama-3.1-70B-Instruct-Q4_K_M.gguf <span class=\"token punctuation\">\\\\<\/span><br \/>\n  &#8211;n-gpu-layers <span class=\"token number\">999<\/span> <span class=\"token punctuation\">\\\\<\/span>          <span class=\"token comment\"># \u5168\u90e8\u5c42\u5378\u8f7d\u5230 GPU<\/span><br \/>\n  <span class=\"token parameter variable\">-fa<\/span> <span class=\"token punctuation\">\\\\<\/span>                          <span class=\"token comment\"># Flash Attention<\/span><br \/>\n  &#8211;ctx-size <span class=\"token number\">32768<\/span> <span class=\"token punctuation\">\\\\<\/span><br \/>\n  <span class=\"token parameter variable\">&#8211;host<\/span> <span class=\"token number\">0.0<\/span>.0.0 <span class=\"token parameter variable\">&#8211;port<\/span> <span class=\"token number\">8080<\/span><\/p>\n<p>Lemonade SDK \u63d0\u4f9b\u4e86\u5185\u7f6e ROCm 7 \u52a0\u901f\u7684 llama.cpp \u591c\u95f4\u6784\u5efa\u7248\u672c&#xff0c;\u4e13\u95e8\u9488\u5bf9 gfx1151&#xff08;Strix Halo \/ Ryzen AI Max&#043; 395&#xff09;\u4f18\u5316\u3002<\/p>\n<p>rocWMMA \u52a0\u901f&#xff08;\u9488\u5bf9 BF16\/FP16 \u5927\u5e45\u63d0\u901f&#xff09;&#xff1a;<\/p>\n<p>rocWMMA \u6807\u8bb0\u7684\u6784\u5efa\u7248\u672c\u5229\u7528\u4e86 AMD \u7684 rocWMMA \u5e93&#xff0c;\u89e3\u9501\u4e86\u901a\u8fc7\u6ce2\u5f62\u77e9\u9635\u4e58\u7d2f\u52a0&#xff08;WMMA&#xff09;\u6307\u4ee4\u52a0\u901f\u7684\u77e9\u9635\u4e58\u6cd5\u6d41\u6c34\u7ebf&#xff0c;\u53ef\u663e\u8457\u63d0\u5347 BF16\/FP16 \u5de5\u4f5c\u8d1f\u8f7d\u7684\u6027\u80fd\u3002<\/p>\n<h4>\u65b9\u6848 C&#xff1a;NPU \u52a0\u901f&#xff08;FastFlowLM&#xff09;<\/h4>\n<p>FastFlowLM \u662f\u4e13\u4e3a AMD Ryzen AI NPU \u8bbe\u8ba1\u7684\u63a8\u7406\u8fd0\u884c\u65f6&#xff0c;\u65e0\u9700 GPU \u5373\u53ef\u8fd0\u884c\u3002\u5b83\u7c7b\u4f3c Ollama&#xff0c;\u4f46\u4e13\u95e8\u9488\u5bf9 NPU \u4f18\u5316&#xff0c;\u652f\u6301\u9ad8\u8fbe 256,000 tokens \u7684\u8d85\u957f\u4e0a\u4e0b\u6587&#xff0c;\u63d0\u4f9b CLI\u3001REST \u548c OpenAI \u517c\u5bb9 API \u63a5\u53e3&#xff0c;\u8fd0\u884c\u65f6\u4ec5\u7ea6 14 MB\u3002<\/p>\n<p><span class=\"token comment\"># \u5b89\u88c5 FastFlowLM&#xff08;\u901a\u8fc7 Lemonade Server&#xff09;<\/span><br \/>\nflm pull qwen3:1.7b<br \/>\nflm run qwen3:1.7b<\/p>\n<p>\u5728 Windows \u4e0b\u7528 NPU \u8fd0\u884c Mistral 7B&#xff0c;\u89e3\u7801\u901f\u5ea6\u7ea6 4\u20135 tok\/s&#xff0c;\u800c\u7eaf GPU \u8def\u5f84\u7ea6 40 tok\/s\u3002AMD \u6b63\u5728\u63a8\u8fdb&#034;\u5206\u89e3\u63a8\u7406&#034;\u6982\u5ff5\u2014\u2014\u8ba1\u7b97\u5bc6\u96c6\u7684\u63d0\u793a\u8bcd\u5904\u7406\u7531 NPU \u627f\u62c5&#xff0c;\u5185\u5b58\u5e26\u5bbd\u5bc6\u96c6\u7684\u89e3\u7801\u9636\u6bb5\u7531 GPU \u5904\u7406&#xff0c;\u8fd9\u5bf9\u529f\u8017\u53d7\u9650\u7684\u7b14\u8bb0\u672c\u7535\u8111\u610f\u4e49\u91cd\u5927\u3002<\/p>\n<hr \/>\n<h3>\u4e94\u3001\u91cf\u5316\u7b56\u7565&#xff1a;\u6a21\u578b\u7cbe\u5ea6\u4e0e\u5185\u5b58\u6743\u8861<\/h3>\n<table>\n<tr>\u91cf\u5316\u683c\u5f0f\u5185\u5b58\u5360\u7528&#xff08;70B&#xff09;\u901f\u5ea6\u8d28\u91cf\u63a8\u8350\u573a\u666f<\/tr>\n<tbody>\n<tr>\n<td>BF16 \u5168\u7cbe\u5ea6<\/td>\n<td>~140 GB<\/td>\n<td>\u6700\u6162<\/td>\n<td>\u6700\u4f18<\/td>\n<td>\u8bc4\u4f30\/\u5fae\u8c03<\/td>\n<\/tr>\n<tr>\n<td>Q8_0<\/td>\n<td>~70 GB<\/td>\n<td>\u8f83\u5feb<\/td>\n<td>\u6781\u4f73<\/td>\n<td>\u9ad8\u8d28\u91cf\u63a8\u7406<\/td>\n<\/tr>\n<tr>\n<td>Q4_K_M<\/td>\n<td>~40 GB<\/td>\n<td>\u5feb<\/td>\n<td>\u826f\u597d<\/td>\n<td>\u65e5\u5e38\u63a8\u7406\u2605<\/td>\n<\/tr>\n<tr>\n<td>Q3_K_M<\/td>\n<td>~30 GB<\/td>\n<td>\u6700\u5feb<\/td>\n<td>\u4e00\u822c<\/td>\n<td>\u8d85\u5927\u6a21\u578b<\/td>\n<\/tr>\n<tr>\n<td>IQ4_XS<\/td>\n<td>~35 GB<\/td>\n<td>\u5feb<\/td>\n<td>\u8f83\u597d<\/td>\n<td>MoE \u6a21\u578b\u2605<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>MoE \u6a21\u578b\u662f Strix Halo \u7684\u9690\u85cf\u738b\u724c\u2014\u2014MoE \u67b6\u6784&#xff08;\u5982 Qwen3 30B-A3B&#xff09;\u867d\u7136\u603b\u53c2\u6570\u91cf\u5927&#xff0c;\u4f46\u6bcf\u6b21\u63a8\u7406\u53ea\u6fc0\u6d3b\u5c11\u91cf\u53c2\u6570&#xff0c;\u5728\u5185\u5b58\u5e26\u5bbd\u53d7\u9650\u7684 APU \u4e0a\u901f\u5ea6\u53cd\u800c\u8fdc\u8d85\u540c\u7b49\u53c2\u6570\u7684\u5bc6\u96c6\u6a21\u578b\u3002<\/p>\n<hr \/>\n<h3>\u516d\u3001\u5b9e\u6d4b\u6027\u80fd\u57fa\u51c6<\/h3>\n<p>\u57fa\u4e8e Vulkan RADV \u540e\u7aef\u7684\u5b9e\u6d4b\u6570\u636e&#xff08;Beelink GTR9 Pro&#xff0c;Ryzen AI Max&#043; 395&#xff0c;128 GB&#xff09;&#xff1a;<\/p>\n<ul>\n<li>Gemma 4 26B-A4B&#xff1a;\u89e3\u7801 48.5 tok\/s<\/li>\n<li>Llama 4 Scout 109B&#xff1a;\u89e3\u7801 18.3 tok\/s&#xff08;RTX 4090 \u6839\u672c\u65e0\u6cd5\u52a0\u8f7d\u6b64\u6a21\u578b&#xff09;<\/li>\n<li>Qwen3.6 MoE&#xff1a;\u7ea6 50.5 tok\/s<\/li>\n<\/ul>\n<hr \/>\n<h3>\u4e03\u3001\u591a\u673a\u96c6\u7fa4&#xff1a;\u7a81\u7834\u5355\u673a\u9650\u5236<\/h3>\n<p>\u5f53\u5355\u53f0\u673a\u5668\u5185\u5b58\u4e0d\u591f\u8dd1\u4e07\u4ebf\u53c2\u6570\u6a21\u578b\u65f6&#xff0c;\u53ef\u901a\u8fc7 llama.cpp \u7684 RPC \u673a\u5236\u7ec4\u7f51&#xff1a;<\/p>\n<p>\u5229\u7528 llama.cpp RPC \u5f15\u64ce&#xff0c;\u4e00\u53f0\u4e3b\u673a\u8d1f\u8d23 tokenization\u3001\u8c03\u5ea6\u548c\u7f16\u6392&#xff0c;\u5176\u4f59\u673a\u5668\u8fd0\u884c\u8f7b\u91cf\u7ea7 RPC server \u66b4\u9732\u672c\u5730 GPU \u5185\u5b58\u548c\u7b97\u529b\u3002\u6a21\u578b\u4ece\u4e3b\u673a\u89c6\u89d2\u770b\u5c31\u50cf\u8fd0\u884c\u5728\u4e00\u5757\u8d85\u5927\u52a0\u901f\u5668\u4e0a&#xff0c;RPC \u5728\u5e55\u540e\u5904\u7406\u5f20\u91cf\u4f20\u8f93\u4e0e\u540c\u6b65\u2014\u2014\u6574\u4e2a\u6846\u67b6\u5b8c\u7f8e\u5339\u914d Ryzen AI Max&#043; \u7684\u7edf\u4e00\u5185\u5b58\u67b6\u6784&#xff0c;\u53ef\u5b9e\u73b0\u5bf9 Kimi K2.5 \u4e07\u4ebf\u53c2\u6570\u6a21\u578b\u7684\u672c\u5730\u63a8\u7406\u3002<\/p>\n<p><span class=\"token comment\"># \u4ece\u8282\u70b9&#xff08;\u6bcf\u53f0\u673a\u5668\u6267\u884c&#xff09;<\/span><br \/>\n.\/rpc-server <span class=\"token parameter variable\">&#8211;host<\/span> <span class=\"token number\">0.0<\/span>.0.0 <span class=\"token parameter variable\">&#8211;port<\/span> <span class=\"token number\">50052<\/span><\/p>\n<p><span class=\"token comment\"># \u4e3b\u8282\u70b9&#xff08;\u5206\u7247\u52a0\u8f7d\u6a21\u578b&#xff09;<\/span><br \/>\n.\/llama-server <span class=\"token punctuation\">\\\\<\/span><br \/>\n  <span class=\"token parameter variable\">-m<\/span> kimi-k2.5-Q4_K_M.gguf <span class=\"token punctuation\">\\\\<\/span><br \/>\n  <span class=\"token parameter variable\">&#8211;rpc<\/span> <span class=\"token number\">192.168<\/span>.1.2:50052,192.168.1.3:50052,192.168.1.4:50052 <span class=\"token punctuation\">\\\\<\/span><br \/>\n  <span class=\"token parameter variable\">-fa<\/span> &#8211;ctx-size <span class=\"token number\">16384<\/span><\/p>\n<hr \/>\n<h3>\u516b\u3001\u5173\u952e\u8c03\u4f18\u6e05\u5355<\/h3>\n<li>\u59cb\u7ec8\u4fdd\u6301 llama.cpp \u6700\u65b0\u7248&#xff1a;\u5bf9 MoE \u6a21\u578b\u800c\u8a00&#xff0c;\u66f4\u65b0 llama.cpp \u5e26\u6765\u7684\u6027\u80fd\u63d0\u5347&#xff08;&#043;25%&#xff09;\u8d85\u8fc7\u4e86\u6240\u6709\u5185\u6838\u53c2\u6570\u8c03\u4f18\u3001\u6279\u6b21\u5927\u5c0f\u8c03\u6574\u548c\u9a71\u52a8\u5bf9\u6bd4\u7684\u603b\u548c\u3002<\/li>\n<li>\u542f\u7528 Flash Attention&#xff1a;-fa 1 \u6216 OLLAMA_FLASH_ATTENTION&#061;1<\/li>\n<li>\u5378\u8f7d AMDVLK&#xff1a;\u9632\u6b62\u9759\u9ed8\u52ab\u6301 Vulkan<\/li>\n<li>\u4f7f\u7528 tuned \u6027\u80fd\u5b88\u62a4\u8fdb\u7a0b&#xff1a;accelerator-performance \u914d\u7f6e\u6587\u4ef6\u53ef\u5e26\u6765 &#043;5\u20138% \u7684 LLM \u901f\u5ea6\u63d0\u5347\u3002<\/li>\n<li>ROCm \u73af\u5883\u53d8\u91cf&#xff1a;HSA_OVERRIDE_GFX_VERSION&#061;11.5.1&#xff08;\u4f7f\u7528 ROCm \u8def\u5f84\u65f6&#xff09;<\/li>\n<p>\u603b\u4f53\u800c\u8a00&#xff0c;Strix Halo \u7684\u6838\u5fc3\u7ade\u4e89\u529b\u5728\u4e8e&#xff1a;\u7528\u7b14\u8bb0\u672c\/\u8ff7\u4f60 PC \u7684\u529f\u8017&#xff08;65\u2013120 W&#xff09;\u548c\u4ef7\u683c&#xff08;\u7ea6 \u00a514,000 \u8d77&#xff09;&#xff0c;\u5b9e\u73b0\u539f\u672c\u9700\u8981\u670d\u52a1\u5668\u7ea7\u591a\u5361\u914d\u7f6e\u624d\u80fd\u652f\u6491\u7684 70B&#043; \u6a21\u578b\u672c\u5730\u63a8\u7406&#xff0c;\u4e14\u6570\u636e\u5b8c\u5168\u4e0d\u51fa\u672c\u5730&#xff0c;\u6ee1\u8db3\u9690\u79c1\u548c\u5408\u89c4\u9700\u6c42\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"<p>\u672c\u6587\u5c06\u7cfb\u7edf\u5730\u4ecb\u7ecd\u4e00\u4e0b\u5982\u4f55\u5728 Strix Halo \u5e73\u53f0\u4e0a\u90e8\u7f72\u670d\u52a1\u5668\u7ea7 AI \u5927\u6a21\u578b\u3002\u5148\u753b\u4e00\u5f20\u6574\u4f53\u67b6\u6784\u56fe&#xff0c;\u518d\u8be6\u7ec6\u5c55\u5f00\u3002\u4e00\u3001\u4e3a\u4ec0\u4e48 Strix Halo \u80fd\u8dd1\u670d\u52a1\u5668\u7ea7\u5927\u6a21\u578b&#xff1f;<br \/>\n\u6838\u5fc3\u4f18\u52bf&#xff1a;\u7edf\u4e00\u5927\u5185\u5b58\u7a81\u7834\u663e\u5b58\u74f6\u9888<br \/>\n\u4f20\u7edf\u6d88\u8d39\u7ea7 GPU&#xff08;RTX 4090&#xff09;\u6700\u591a\u53ea\u6709 24 GB \u72ec\u7acb\u663e\u5b58&#xff0c;\u65e0\u6cd5\u5728\u5355\u5361\u4e0a\u8fd0\u884c 70B \u53c2\u6570\u6a21\u578b\u3002Strix Halo&#xff08;Ryzen AI Max 395&#xff09;<\/p>\n","protected":false},"author":2,"featured_media":82093,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[208,50,188],"topic":[],"class_list":["post-82094","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-server","tag-gpu","tag-50","tag-188"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/82094.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u672c\u6587\u5c06\u7cfb\u7edf\u5730\u4ecb\u7ecd\u4e00\u4e0b\u5982\u4f55\u5728 Strix Halo \u5e73\u53f0\u4e0a\u90e8\u7f72\u670d\u52a1\u5668\u7ea7 AI \u5927\u6a21\u578b\u3002\u5148\u753b\u4e00\u5f20\u6574\u4f53\u67b6\u6784\u56fe&#xff0c;\u518d\u8be6\u7ec6\u5c55\u5f00\u3002\u4e00\u3001\u4e3a\u4ec0\u4e48 Strix Halo \u80fd\u8dd1\u670d\u52a1\u5668\u7ea7\u5927\u6a21\u578b&#xff1f; \u6838\u5fc3\u4f18\u52bf&#xff1a;\u7edf\u4e00\u5927\u5185\u5b58\u7a81\u7834\u663e\u5b58\u74f6\u9888 \u4f20\u7edf\u6d88\u8d39\u7ea7 GPU&#xff08;RTX 4090&#xff09;\u6700\u591a\u53ea\u6709 24 GB \u72ec\u7acb\u663e\u5b58&#xff0c;\u65e0\u6cd5\u5728\u5355\u5361\u4e0a\u8fd0\u884c 70B \u53c2\u6570\u6a21\u578b\u3002Strix Halo&#xff08;Ryzen AI Max 395&#xff09;\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/82094.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-07-24T22:36:45+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/07\/20260724223643-6a63e8fb7f4c9.png\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"3 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/82094.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/82094.html\",\"name\":\"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-07-24T22:36:45+00:00\",\"dateModified\":\"2026-07-24T22:36:45+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/82094.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/82094.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/82094.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/82094.html","og_locale":"zh_CN","og_type":"article","og_title":"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u672c\u6587\u5c06\u7cfb\u7edf\u5730\u4ecb\u7ecd\u4e00\u4e0b\u5982\u4f55\u5728 Strix Halo \u5e73\u53f0\u4e0a\u90e8\u7f72\u670d\u52a1\u5668\u7ea7 AI \u5927\u6a21\u578b\u3002\u5148\u753b\u4e00\u5f20\u6574\u4f53\u67b6\u6784\u56fe&#xff0c;\u518d\u8be6\u7ec6\u5c55\u5f00\u3002\u4e00\u3001\u4e3a\u4ec0\u4e48 Strix Halo \u80fd\u8dd1\u670d\u52a1\u5668\u7ea7\u5927\u6a21\u578b&#xff1f; \u6838\u5fc3\u4f18\u52bf&#xff1a;\u7edf\u4e00\u5927\u5185\u5b58\u7a81\u7834\u663e\u5b58\u74f6\u9888 \u4f20\u7edf\u6d88\u8d39\u7ea7 GPU&#xff08;RTX 4090&#xff09;\u6700\u591a\u53ea\u6709 24 GB \u72ec\u7acb\u663e\u5b58&#xff0c;\u65e0\u6cd5\u5728\u5355\u5361\u4e0a\u8fd0\u884c 70B \u53c2\u6570\u6a21\u578b\u3002Strix Halo&#xff08;Ryzen AI Max 395&#xff09;","og_url":"https:\/\/www.wsisp.com\/helps\/82094.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-07-24T22:36:45+00:00","og_image":[{"url":"https:\/\/www.wsisp.com\/helps\/wp-content\/uploads\/2026\/07\/20260724223643-6a63e8fb7f4c9.png"}],"author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"3 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/82094.html","url":"https:\/\/www.wsisp.com\/helps\/82094.html","name":"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-07-24T22:36:45+00:00","dateModified":"2026-07-24T22:36:45+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/82094.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/82094.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/82094.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"AMD Ryzen AI Strix Halo\u67b6\u6784\u5904\u7406\u5668\uff1a\u5982\u4f55\u5728\u7b14\u8bb0\u672c\u4e0a\u8dd1\u901a\u539f\u672c\u5c5e\u4e8e\u670d\u52a1\u5668\u7684\u6a21\u578b\uff1f"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/82094","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=82094"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/82094\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media\/82093"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=82094"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=82094"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=82094"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=82094"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}