{"id":86207,"topic":"ai","source":"Sohu","title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","url_hash":"5bb10332fb75ff23ebc61cd290181bb4fab1b821","author":"","summary":"<a href=\"https://news.google.com/rss/articles/CBMijAFBVV95cUxPVS1HRXd0amtBM18tQU5qM3Q1Z19YUGVEcnVqXzdRT0V0Vy1jZVFPWGhSQjYtLUkzUkt5b19acnJLT19vcWpOUEJJZVEtSko1b0JZTzctV0xWWnQ0YjhsX1FodmkzamJKMGRhb042WWVud01NYmw4eDBuT2dEcWduQXhiVUF0QmJ3Szk0VA?oc=5\" target=\"_blank\">实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">Sohu</font>","content":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","image_url":"//q5.itc.cn/images01/20260920/1dbf61a4257a435bb70c6f9f4355706c.jpeg","lang":"zh","published_at":"2026-09-20T03:17:34+00:00","fetched_at":"2026-09-20T04:15:08+00:00","status":"read","starred":0,"extract_state":"ok","summary_auto":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","cluster_id":null,"extract_retries":0,"extract_error":null,"contract_version":"news_item.v1","format_contract_version":"news_item_formats.v1","dedup_url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 651 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":651,"summary_length":651,"usable_text_length":651,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":651,"summary_length":651}},"news_item":{"id":86207,"canonical_url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","source_url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","source_name":"Sohu","author":null,"published_at":"2026-09-20T03:17:34+00:00","locale":"zh","topic":"ai","tags":[],"rss_summary":"<a href=\"https://news.google.com/rss/articles/CBMijAFBVV95cUxPVS1HRXd0amtBM18tQU5qM3Q1Z19YUGVEcnVqXzdRT0V0Vy1jZVFPWGhSQjYtLUkzUkt5b19acnJLT19vcWpOUEJJZVEtSko1b0JZTzctV0xWWnQ0YjhsX1FodmkzamJKMGRhb042WWVud01NYmw4eDBuT2dEcWduQXhiVUF0QmJ3Szk0VA?oc=5\" target=\"_blank\">实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香</a>&nbsp;&nbsp;<font color=\"#6f6f6f\">Sohu</font>","full_text":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","excerpt":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 651 characters.","diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 651 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":651,"summary_length":651,"usable_text_length":651,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":651,"summary_length":651}}},"display_formats":["compact","card","full","digest_section","json"]},"daily_stack_record":{"title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","summary":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","source":"Sohu","date":"2026-09-20T03:17:34+00:00","content":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 651 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 651 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":651,"summary_length":651,"usable_text_length":651,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":651,"summary_length":651}},"tags":[]},"fallback_formats":["markdown","json","html"],"actions":{"read":"/item/86207","export_markdown":"/api/items/86207/export?format=markdown","export_json":"/api/items/86207/export?format=json","diagnose":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334"},"formats":{"full":{"id":86207,"title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","source":"Sohu","author":null,"published_at":"2026-09-20T03:17:34+00:00","locale":"zh","topic":"ai","tags":[],"excerpt":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","full_text":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","reading_time_min":1,"extraction":{"state":"ok","confidence":0.9,"error":null,"explanation":"High confidence: full text extraction produced 651 characters.","diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 651 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":651,"summary_length":651,"usable_text_length":651,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":651,"summary_length":651}}},"quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 651 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":651,"summary_length":651,"usable_text_length":651,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":651,"summary_length":651}},"actions":{"read":"/item/86207","export_markdown":"/api/items/86207/export?format=markdown","export_json":"/api/items/86207/export?format=json","diagnose":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334"}},"digest":{"id":86207,"title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","source":"Sohu","topic":"ai","published_at":"2026-09-20T03:17:34+00:00","excerpt":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。 一、先搞清楚：AI PC到底加了啥 AI…","quality_bucket":"high","quality_reason":"High confidence: full text extraction produced 651 characters.","reading_time_min":1,"cluster_id":null},"card":{"display_title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","subtitle":"Sohu · 2026-09-20","summary":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。 一、先搞清楚：AI PC到底加了啥 AI…","badges":["quality:high"],"links":{"read":"/item/86207","original":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","diagnose":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334"},"quality_warning":null},"export":{"title":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香 - Sohu","url":"https://m.sohu.com/a/1078494944_121761735?scm=10001.325_13-325_13.0.0-0-0-0-0.5_1334","summary":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","source":"Sohu","date":"2026-09-20T03:17:34+00:00","content":"实测3台热门AI PC：本地大模型“能跑不等于能用”，这些功能才是真香\n这两年“AI PC”三个字被吹上天：NPU算力从40 TOPS卷到80 TOPS，厂商恨不得每台电脑都叫AI PC。但网上争议也大，有人说真好用，有人说纯割韭菜。我最近把三台热门AI PC的实测数据从头到尾过了一遍，也翻了大量网友的真实反馈，哪些是营销话术，哪些真能落地，今天一次说透。\n一、先搞清楚：AI PC到底加了啥\nAI PC的核心就一个字：NPU，也就是CPU里塞一块专门跑AI的芯片。它的优点是省电——按英特尔白皮书的数据，NPU功耗是毫瓦级，CPU和GPU都是瓦级。AMD今年新平台最高给到60 TOPS，高通骁龙X2 Elite甚至标到80 TOPS。但注意：TOPS是算力上限，不是应用数量。很多软件根本不用本地NPU，直接调云端接口，这就是为什么有网友吐槽“任务管理器里NPU常年归零”。\n二、本地大模型：能跑不等于能用\n先说结论：本地大模型“能跑”和“能用”是两回事。拿AMD锐龙AI 9 HX 470这类旗舰平台实测：Nemotron 4B小模型每秒21个token左右，DeepSeek-R1 8B每秒17个，Qwen3.5 9B只剩12个。什么概念？8B模型生成一段300字回复要半分钟，复杂推理直接露馅。而且16G内存跑不了多大模型，32G才刚起步。网友总结得很扎心：“同样的提问，云端豆包、GPT回答又快又全面，谁没事折腾本地部署？”本地大模型目前只适合两类人：数据不能出内网的办公用户，和折腾党。","confidence":0.9,"diagnostics_url":"/api/diagnose?url=https%3A//m.sohu.com/a/1078494944_121761735%3Fscm%3D10001.325_13-325_13.0.0-0-0-0-0.5_1334","quality_bucket":"high","failure_kind":"none","retryable":false,"quality_reason":"High confidence: full text extraction produced 651 characters.","quality_profile":{"profile_version":"extraction_quality.v2","bucket":"high","confidence":0.9,"failure_kind":"none","retryable":false,"retry_after_attempts":0,"reason":"High confidence: full text extraction produced 651 characters.","operator_guidance":{"severity":"ok","recommended_action":"trust_full_text","next_step":"Use the extracted full text as the primary article source.","operator_label":"Ready","can_retry":false,"can_use_summary":false,"diagnostics_required":false},"content_depth":{"contract_version":"content_depth.v1","category":"full_text","label":"Full text","has_full_text":true,"has_summary":true,"content_length":651,"summary_length":651,"usable_text_length":651,"source_field":"content"},"legacy_collapsed":false,"signals":{"extract_state":"ok","extract_error":null,"extract_retries":0,"content_length":651,"summary_length":651}},"tags":[],"format_contract_version":"news_item_formats.v1"}}}