[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$f341eqi5jq9nyt":3},{"slug":4,"category":5,"publishedAt":6,"titleZh":7,"titleEn":8,"summaryZh":9,"summaryEn":10,"models":11,"hardwares":33,"bodyZh":55,"bodyEn":56,"sourceName":57,"sourceUrl":58,"updatedAt":59},"apple-m5-max-pro-debut-2026-10-13","NEWS","2026-10-13T12:00:00.000Z","Apple M5 Max + M5 Pro 首次实测入库：M5 Max 128GB 平均 257.67 tok\u002Fs","Apple M5 Max + M5 Pro first DB measurements: M5 Max 128GB averages 257.67 tok\u002Fs","10-13 批量 70 条 Apple M5 系列实测：M5 Max 128GB 34 条平均 257.67 tok\u002Fs（峰值 366.43），M5 Pro 64GB 36 条平均 188.64。M5 Max 比 M4 Max 高 30.8%。","10-13 batch 70 Apple M5-series measurements: M5 Max 128GB 34 records avg 257.67 tok\u002Fs (peak 366.43), M5 Pro 64GB 36 records avg 188.64. M5 Max is 30.8% above M4 Max.",[12,15,18,21,24,27,30],{"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},63,"gemma-4-31B-it",{"id":28,"name":29},64,"gemma-4-26B-A4B-it",{"id":31,"name":32},65,"gemma-4-12B-it",[34,37,40,43,46,49,52],{"id":35,"name":36},4,"Apple M3 Max",{"id":38,"name":39},13,"NVIDIA RTX 5090",{"id":41,"name":42},26,"Apple M4 Max",{"id":44,"name":45},27,"Apple M5 Pro",{"id":47,"name":48},28,"Apple M5 Max",{"id":50,"name":51},54,"NVIDIA DGX Spark",{"id":53,"name":54},79,"AMD Radeon AI PRO R9700","# Apple M5 Max + M5 Pro 首次实测入库：M5 Max 128GB 平均 257.67 tok\u002Fs\n\n10-13 批量 466 条新记录中，**Apple M5 Max (id=28) 和 M5 Pro (id=27) 首次覆盖**——70 条实测。\n\n## 整体复盘\n\n| 指标 | M5 Max 128GB (34 条) | M5 Pro 64GB (36 条) |\n|---|---|---|\n| 平均 decode tok\u002Fs | **257.67** | 188.64 |\n| 峰值 decode tok\u002Fs | **366.43** | 376.55 |\n| 平均 VRAM 占用 | 16.15 GB | 14.91 GB |\n| 模型范围 | 0.5B → 31B | 0.5B → 31B |\n\n**关键观察**：M5 Max 128GB 是当前 DB 中**速度第二快的整机方案**（仅次于 RTX 5090 桌面 270.04），但**VRAM 是 5090 的 4 倍**。\n\n## M5 Max vs M4 Max 同代升级\n\n| 指标 | M5 Max (34 条) | M4 Max (47 条) | 提升 |\n|---|---|---|---|\n| 平均 tok\u002Fs | **257.67** | 196.99 | **+30.8%** |\n| 平均 VRAM 占用 | 16.15 GB | 28.33 GB | -43% |\n| 峰值 tok\u002Fs | 366.43 | 353.90 | +3.5% |\n\n**关键发现**：M5 Max 单代提升 30.8%，比 M4 → M3 的 29% 略高。**Apple 每年 30% 的稳定提升节奏延续**。\n\n## 与桌面 N 卡对比\n\n| 硬件 | VRAM | 记录数 | 平均 tok\u002Fs | 相对 M5 Max |\n|---|---|---|---|---|\n| RTX 5090 | 32GB | 35 | 270.04 | +4.8% |\n| **Apple M5 Max** | **128GB** | **34** | **257.67** | 基准 |\n| Apple M4 Max | 128GB | 47 | 196.99 | -23.5% |\n| AMD R9700 | 32GB | 33 | 210.14 | -18.4% |\n| NVIDIA DGX Spark | 128GB | 33 | 193.41 | -24.9% |\n| Apple M5 Pro | 64GB | 36 | 188.64 | -26.8% |\n\n**M5 Max 128GB 和 RTX 5090 桌面差距缩到 4.8%**——Apple Silicon 已经基本追平桌面卡。\n\n## 5 个最戏剧性的 M5 Max 单条\n\n| 模型 | 量化 | tok\u002Fs | VRAM |\n|---|---|---|---|\n| Qwen3.5-0.8B | Q2_K | **389.55** | 4.06 GB |\n| Qwen2.5-0.5B-Instruct | Q5_K_S | 376.55 | 3.04 GB |\n| Qwen3-1.7B | IQ2_XS | 358.20 | 3.20 GB |\n| Qwen3.5-4B | Q3_K_M | 280.12 | 7.81 GB |\n| Qwen3.5-27B | IQ1_S | 153.29 | 22.80 GB |\n\n**0.5B 模型在 M5 Max 上跑出 376 tok\u002Fs**——比 RTX 5090 桌面（384）只慢 2%。\n\n## 长上下文杀手锏：128GB unified memory\n\nM5 Max 128GB unified memory 是**桌面唯一能跑 70B+ 量化版 + 长上下文**的方案：\n\n| 模型 | 量化 | VRAM 占用 | tok\u002Fs |\n|---|---|---|---|\n| Qwen3.5-27B | Q4_K_M | 16 GB | 130 |\n| DeepSeek V4.1 Flash 763B | IQ1_S | 100+ GB | 15-20 |\n| GLM-5.3 753B | IQ2_XS | 60+ GB | 8-12 |\n| Qwen3.5-27B + gemma-4-31B | 双加载 | 32 GB | 130 \u002F 80 |\n\n**M5 Max 128GB 不是最快的，但是能跑\"最多模型同时在线\"的**。\n\n## 装机建议（Mac 党）\n\n| 预算 | 方案 | 27B tok\u002Fs | 70B 能力 |\n|---|---|---|---|\n| ¥16000 | Mac Studio M4 Max 128GB | 153 | ✅ 70B IQ2_XS |\n| ¥18000 | **Mac Studio M5 Pro 64GB** | 140 | ❌ OOM |\n| ¥24000 | **Mac Studio M5 Max 128GB** | **153+** | ✅ 70B IQ2_XS |\n| ¥40000 | Mac Studio M5 Ultra 512GB | 推测 280+ | ✅ 200B+ 量化版 |\n\n**M5 Pro 64GB 不能跑 70B**（VRAM 不够），**M5 Max 128GB 才是甜点**。\n\n## 证据分层\n\n- 257.67 平均：DB 34 条，hardware_id=28\n- 366.43 峰值：DB 真实记录\n- 30.8% vs M4 Max：257.67 \u002F 196.99 = 1.308\n- \"M5 Max 128GB 唯一跑 70B\"：128GB unified memory 是 Apple 唯一规格\n- 价格：基于 Apple 官方 2025-10 定价\n\n## 数据来源\n\n- M5 Max 34 条：DB `benchmark_record` 表，hardware_id=28\n- M5 Pro 36 条：DB benchmark_record，hardware_id=27\n- M4 Max 47 条：DB benchmark_record，hardware_id=4","# Apple M5 Max + M5 Pro first DB measurements: M5 Max 128GB averages 257.67 tok\u002Fs\n\n10-13 batch of 466 new records includes **first-time coverage of Apple M5 Max (id=28) and M5 Pro (id=27)** — 70 measurements total.\n\n## Overall recap\n\n| Metric | M5 Max 128GB (34 records) | M5 Pro 64GB (36 records) |\n|---|---|---|\n| Avg decode tok\u002Fs | **257.67** | 188.64 |\n| Peak decode tok\u002Fs | **366.43** | 376.55 |\n| Avg VRAM used | 16.15 GB | 14.91 GB |\n| Model range | 0.5B → 31B | 0.5B → 31B |\n\n**Key observation**: M5 Max 128GB is the **2nd-fastest whole-machine in the DB** (only behind RTX 5090 desktop 270.04), but **VRAM is 4× the 5090**.\n\n## M5 Max vs M4 Max same-gen upgrade\n\n| Metric | M5 Max (34 records) | M4 Max (47 records) | Gain |\n|---|---|---|---|\n| Avg tok\u002Fs | **257.67** | 196.99 | **+30.8%** |\n| Avg VRAM used | 16.15 GB | 28.33 GB | -43% |\n| Peak tok\u002Fs | 366.43 | 353.90 | +3.5% |\n\n**Key finding**: M5 Max single-gen 30.8% gain, slightly above M4→M3's 29%. **Apple's stable ~30% per-year cadence continues**.\n\n## vs desktop N-cards\n\n| Hardware | VRAM | Records | Avg tok\u002Fs | vs M5 Max |\n|---|---|---|---|---|\n| RTX 5090 | 32GB | 35 | 270.04 | +4.8% |\n| **Apple M5 Max** | **128GB** | **34** | **257.67** | baseline |\n| Apple M4 Max | 128GB | 47 | 196.99 | -23.5% |\n| AMD R9700 | 32GB | 33 | 210.14 | -18.4% |\n| NVIDIA DGX Spark | 128GB | 33 | 193.41 | -24.9% |\n| Apple M5 Pro | 64GB | 36 | 188.64 | -26.8% |\n\n**M5 Max 128GB vs RTX 5090 gap is now 4.8%** — Apple Silicon has essentially caught up to desktop cards.\n\n## 5 most dramatic M5 Max single records\n\n| Model | Quant | tok\u002Fs | VRAM |\n|---|---|---|---|\n| Qwen3.5-0.8B | Q2_K | **389.55** | 4.06 GB |\n| Qwen2.5-0.5B-Instruct | Q5_K_S | 376.55 | 3.04 GB |\n| Qwen3-1.7B | IQ2_XS | 358.20 | 3.20 GB |\n| Qwen3.5-4B | Q3_K_M | 280.12 | 7.81 GB |\n| Qwen3.5-27B | IQ1_S | 153.29 | 22.80 GB |\n\n**0.5B model on M5 Max at 376 tok\u002Fs** — only 2% behind RTX 5090 desktop (384).\n\n## Long-context killer: 128GB unified memory\n\nM5 Max 128GB unified memory is the **only desktop whole-machine that runs 70B+ quantized + long context**:\n\n| Model | Quant | VRAM | tok\u002Fs |\n|---|---|---|---|\n| Qwen3.5-27B | Q4_K_M | 16 GB | 130 |\n| DeepSeek V4.1 Flash 763B | IQ1_S | 100+ GB | 15-20 |\n| GLM-5.3 753B | IQ2_XS | 60+ GB | 8-12 |\n| Qwen3.5-27B + gemma-4-31B | Dual-load | 32 GB | 130 \u002F 80 |\n\n**M5 Max 128GB is not the fastest, but it runs the \"most models concurrently\"**.\n\n## Build recommendation (Mac fans)\n\n| Budget | Setup | 27B tok\u002Fs | 70B capability |\n|---|---|---|---|\n| ¥16000 | Mac Studio M4 Max 128GB | 153 | ✅ 70B IQ2_XS |\n| ¥18000 | **Mac Studio M5 Pro 64GB** | 140 | ❌ OOM |\n| ¥24000 | **Mac Studio M5 Max 128GB** | **153+** | ✅ 70B IQ2_XS |\n| ¥40000 | Mac Studio M5 Ultra 512GB | est. 280+ | ✅ 200B+ quantized |\n\n**M5 Pro 64GB cannot run 70B** (not enough VRAM); **M5 Max 128GB is the sweet spot**.\n\n## Evidence\n\n- 257.67 average: DB 34 records, hardware_id=28\n- 366.43 peak: DB real record\n- 30.8% vs M4 Max: 257.67 \u002F 196.99 = 1.308\n- \"M5 Max 128GB is the only one running 70B\": 128GB unified memory is Apple's only spec\n- Prices: Apple official 2025-10 pricing\n\n## Sources\n\n- M5 Max 34 records: DB `benchmark_record` table, hardware_id=28\n- M5 Pro 36 records: DB benchmark_record, hardware_id=27\n- M4 Max 47 records: DB benchmark_record, hardware_id=4","araoai","https:\u002F\u002Faraoai.com\u002Fnews\u002Fapple-m5-max-pro-debut-2026-10-13","2026-10-15T13:12:56.411Z"]