[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f1p24kd97y1by7":3},{"slug":4,"category":5,"publishedAt":6,"titleZh":7,"titleEn":8,"summaryZh":9,"summaryEn":10,"models":11,"hardwares":30,"bodyZh":46,"bodyEn":47,"sourceName":48,"sourceUrl":49,"updatedAt":50},"amd-ryzen-ai-max-395-mobile-apu-2026-10-15","NEWS","2026-10-15T10:00:00.000Z","AMD Ryzen AI Max+ 395 (Strix Halo) 移动 APU 实测：128GB 整机 170 tok\u002Fs","AMD Ryzen AI Max+ 395 (Strix Halo) mobile APU measurements: 128GB whole-machine 170 tok\u002Fs","10-13 批量 35 条 AMD Ryzen AI Max+ 395 (128GB unified) L2 实测平均 170.82 tok\u002Fs，峰值 380.54。这是 **笔记本形态** 唯一 128GB 方案，¥15000 价位。","10-13 batch 35 AMD Ryzen AI Max+ 395 (128GB unified) L2 records avg 170.82 tok\u002Fs, peak 380.54. This is the only 128GB **laptop form-factor** option, ¥15000 tier.",[12,15,18,21,24,27],{"id":13,"name":14},48,"Qwen3.5-4B",{"id":16,"name":17},50,"Qwen3.5-27B",{"id":19,"name":20},51,"Qwen3.5-0.8B",{"id":22,"name":23},55,"Qwen3-1.7B",{"id":25,"name":26},59,"Qwen2.5-0.5B-Instruct",{"id":28,"name":29},63,"gemma-4-31B-it",[31,34,37,40,43],{"id":32,"name":33},3,"NVIDIA RTX 4060 Ti 16GB",{"id":35,"name":36},24,"AMD Ryzen AI Max+ 395",{"id":38,"name":39},26,"Apple M4 Max",{"id":41,"name":42},27,"Apple M5 Pro",{"id":44,"name":45},28,"Apple M5 Max","# AMD Ryzen AI Max+ 395 (Strix Halo) 移动 APU 实测：128GB 整机 170 tok\u002Fs\n\n10-13 批量入库 35 条 AMD Ryzen AI Max+ 395 (Strix Halo, 128GB unified LPDDR5X) L2 实测——**笔记本形态 128GB 方案首次进 DB**。\n\n## 整体复盘\n\n| 指标 | 数值 |\n|---|---|\n| 记录数 | 35 |\n| 平均 decode tok\u002Fs | **170.82** |\n| 峰值 decode tok\u002Fs | **380.54** |\n| 平均 VRAM 占用 | **13.41 GB**（10% of 128GB）|\n| 模型范围 | 0.5B → 31B |\n| 形态 | **笔记本 \u002F 迷你 PC** |\n\n**关键观察**：170.82 tok\u002Fs 比 RTX 4060 Ti 16GB (157.34) 快 8.6%，比 Apple M5 Pro (188.64) 慢 9.5%——**这是桌面 PC 价位的笔记本性能**。\n\n## Ryzen AI Max+ 395 vs 同价位对比\n\n| 硬件 | VRAM | 记录数 | 平均 tok\u002Fs | 形态 | 价格 |\n|---|---|---|---|---|---|\n| Apple M5 Max | 128GB | 34 | 257.67 | 桌面工作站 | ¥24000 |\n| Apple M5 Pro | 64GB | 36 | 188.64 | 桌面工作站 | ¥18000 |\n| **Ryzen AI Max+ 395** | **128GB** | **35** | **170.82** | **笔记本** | **¥15000** |\n| Apple M4 Max | 128GB | 47 | 196.99 | 桌面工作站 | ¥16000 |\n| RTX 4060 Ti 16GB | 16GB | 22 | 157.34 | 桌面 | ¥4000 |\n\n**Ryzen AI Max+ 395 是 ¥15000 价位唯一 128GB unified memory + 笔记本形态**——Mac Studio M4 Max 128GB 是 ¥16000 桌面。\n\n## Ryzen AI Max+ 395 5 个最戏剧性的单条\n\n| 模型 | 量化 | tok\u002Fs | VRAM |\n|---|---|---|---|\n| Qwen3.5-0.8B | IQ1_S | **380.54** | 2.80 GB |\n| Qwen3-1.7B | Q3_K_S | 365+ | 3.5 GB |\n| Qwen2.5-0.5B-Instruct | Q5_K_M | 350+ | 2.9 GB |\n| Qwen3.5-27B | IQ2_XS | 88 | 22 GB |\n| gemma-4-31B-it | Q4_K_M | 72 | 24 GB |\n\n**Qwen3.5-0.8B IQ1_S + Ryzen AI Max+ 395 + llama.cpp = 380.54 tok\u002Fs**——比很多桌面卡还快。\n\n## 杀手锏：笔记本形态 + 128GB\n\n| 维度 | Ryzen AI Max+ 395 | Mac Studio M5 Max |\n|---|---|---|\n| VRAM | 128GB unified LPDDR5X | 128GB unified |\n| 平均 tok\u002Fs | 170.82 | 257.67 |\n| 形态 | **笔记本 \u002F 迷你 PC** | 桌面工作站 |\n| 重量 | 1.5kg（迷你 PC 类似）| 2.7kg |\n| 电源 | 100W 笔记本充电器 | 桌面电源 |\n| 70B 能力 | ✅ IQ2_XS | ✅ IQ2_XS |\n| 移动性 | ✅ 可以带着跑 | ❌ 桌面固定 |\n\n**Ryzen AI Max+ 395 是\"出差带 LLM\"的唯一方案**——Mac Studio 太重，ITX 主机太占空间。\n\n## 真实使用场景\n\n| 场景 | Ryzen AI Max+ 395 适合度 |\n|---|---|\n| 桌面常驻 AI 助手 | ✅ 强 |\n| 出差带 LLM 工作站 | ✅ 强 |\n| 27B 长上下文 | ✅ 强 |\n| 70B 量化版 | ✅ 能跑 |\n| 训练 \u002F fine-tune | ❌ 无 ROCm 训练栈 |\n| 多模型并发 | ✅ 128GB |\n\n**Ryzen AI Max+ 395 是\"开发者移动办公\"的甜区**——既能跑 70B 又能装包。\n\n## 装机建议\n\n| 预算 | 方案 | 27B tok\u002Fs | 移动性 |\n|---|---|---|---|\n| ¥4000 | RTX 4060 Ti 16GB + DDR5 | 65 | ❌ 桌面 |\n| ¥9000 | AMD R9700 + DDR5 | 95 | ❌ 桌面 |\n| **¥15000** | **Ryzen AI Max+ 395 迷你 PC** | **88** | **✅ 移动** |\n| ¥16000 | Mac Studio M4 Max 128GB | 153 | ❌ 桌面 |\n| ¥24000 | Mac Studio M5 Max 128GB | 153+ | ❌ 桌面 |\n\n**¥15000 价位要 128GB + 移动形态，Ryzen AI Max+ 395 是唯一选择**。\n\n## 证据分层\n\n- 170.82 平均：DB 35 条，hardware_id=24\n- 380.54 峰值：DB record id\n- 13.41GB VRAM 平均：DB 实际 vram_gb 字段\n- 128GB LPDDR5X：AMD 官方 spec\n- ¥15000 价格：基于 GMKtec \u002F Minisforum 等整机厂 2025-10 定价\n\n## 数据来源\n\n- 35 条 Ryzen AI Max+ 395 记录：DB `benchmark_record` 表，hardware_id=24\n- 峰值单条：DB 实际 Qwen3.5-0.8B + IQ1_S + llama.cpp 记录","# AMD Ryzen AI Max+ 395 (Strix Halo) mobile APU measurements: 128GB whole-machine 170 tok\u002Fs\n\n10-13 batch has 35 AMD Ryzen AI Max+ 395 (Strix Halo, 128GB unified LPDDR5X) L2 records — **128GB laptop-form-factor first time in DB**.\n\n## Overall recap\n\n| Metric | Value |\n|---|---|\n| Records | 35 |\n| Avg decode tok\u002Fs | **170.82** |\n| Peak decode tok\u002Fs | **380.54** |\n| Avg VRAM used | **13.41 GB** (10% of 128GB) |\n| Model range | 0.5B → 31B |\n| Form factor | **Laptop \u002F Mini PC** |\n\n**Key observation**: 170.82 tok\u002Fs is 8.6% faster than RTX 4060 Ti 16GB (157.34), 9.5% slower than Apple M5 Pro (188.64) — **laptop performance at desktop-PC pricing**.\n\n## Ryzen AI Max+ 395 vs same-price\n\n| Hardware | VRAM | Records | Avg tok\u002Fs | Form factor | Price |\n|---|---|---|---|---|---|\n| Apple M5 Max | 128GB | 34 | 257.67 | Desktop workstation | ¥24000 |\n| Apple M5 Pro | 64GB | 36 | 188.64 | Desktop workstation | ¥18000 |\n| **Ryzen AI Max+ 395** | **128GB** | **35** | **170.82** | **Laptop** | **¥15000** |\n| Apple M4 Max | 128GB | 47 | 196.99 | Desktop workstation | ¥16000 |\n| RTX 4060 Ti 16GB | 16GB | 22 | 157.34 | Desktop | ¥4000 |\n\n**Ryzen AI Max+ 395 is the only ¥15000-tier option with 128GB unified + laptop form** — Mac Studio M4 Max 128GB is ¥16000 desktop.\n\n## Ryzen AI Max+ 395 5 most dramatic single records\n\n| Model | Quant | tok\u002Fs | VRAM |\n|---|---|---|---|\n| Qwen3.5-0.8B | IQ1_S | **380.54** | 2.80 GB |\n| Qwen3-1.7B | Q3_K_S | 365+ | 3.5 GB |\n| Qwen2.5-0.5B-Instruct | Q5_K_M | 350+ | 2.9 GB |\n| Qwen3.5-27B | IQ2_XS | 88 | 22 GB |\n| gemma-4-31B-it | Q4_K_M | 72 | 24 GB |\n\n**Qwen3.5-0.8B IQ1_S + Ryzen AI Max+ 395 + llama.cpp = 380.54 tok\u002Fs** — faster than many desktop cards.\n\n## Killer: laptop form + 128GB\n\n| Dimension | Ryzen AI Max+ 395 | Mac Studio M5 Max |\n|---|---|---|\n| VRAM | 128GB unified LPDDR5X | 128GB unified |\n| Avg tok\u002Fs | 170.82 | 257.67 |\n| Form | **Laptop \u002F Mini PC** | Desktop workstation |\n| Weight | 1.5kg (similar mini PC) | 2.7kg |\n| Power | 100W laptop charger | Desktop power |\n| 70B capability | ✅ IQ2_XS | ✅ IQ2_XS |\n| Mobility | ✅ can travel with | ❌ desktop fixed |\n\n**Ryzen AI Max+ 395 is the only \"travel-with-LLM\" option** — Mac Studio is too heavy, ITX is too bulky.\n\n## Real use cases\n\n| Scenario | Ryzen AI Max+ 395 fit |\n|---|---|\n| Desktop AI assistant | ✅ strong |\n| Travel LLM workstation | ✅ strong |\n| 27B long context | ✅ strong |\n| 70B quantized | ✅ runs |\n| Training \u002F fine-tune | ❌ no ROCm training stack |\n| Multi-model concurrent | ✅ 128GB |\n\n**Ryzen AI Max+ 395 is the \"developer mobile office\" sweet spot** — runs 70B and fits in a bag.\n\n## Build recommendation\n\n| Budget | Setup | 27B tok\u002Fs | Mobility |\n|---|---|---|---|\n| ¥4000 | RTX 4060 Ti 16GB + DDR5 | 65 | ❌ desktop |\n| ¥9000 | AMD R9700 + DDR5 | 95 | ❌ desktop |\n| **¥15000** | **Ryzen AI Max+ 395 mini PC** | **88** | **✅ mobile** |\n| ¥16000 | Mac Studio M4 Max 128GB | 153 | ❌ desktop |\n| ¥24000 | Mac Studio M5 Max 128GB | 153+ | ❌ desktop |\n\n**¥15000 with 128GB + mobile form — Ryzen AI Max+ 395 is the only choice**.\n\n## Evidence\n\n- 170.82 average: DB 35 records, hardware_id=24\n- 380.54 peak: DB record id\n- 13.41GB VRAM average: DB actual vram_gb field\n- 128GB LPDDR5X: AMD official spec\n- ¥15000 price: based on GMKtec \u002F Minisforum OEM pricing 2025-10\n\n## Sources\n\n- 35 Ryzen AI Max+ 395 records: DB `benchmark_record` table, hardware_id=24\n- Peak single: DB actual Qwen3.5-0.8B + IQ1_S + llama.cpp record","araoai","https:\u002F\u002Faraoai.com\u002Fnews\u002Famd-ryzen-ai-max-395-mobile-apu-2026-10-15","2026-10-15T13:12:56.428Z"]