--- title: Running local AI models on Microsoft’s Surface Laptop... description: Hands-on testing of Microsoft's Surface Laptop Ultra (Nvidia N1X Blackwell GPU, up to 128GB unified memory) for running local AI models. On the $4,300 64GB... canonical: https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Running local AI models on Microsoft’s Surface Laptop Ultra: What we found | daily.dev og:description: Hands-on testing of Microsoft's Surface Laptop Ultra (Nvidia N1X Blackwell GPU, up to 128GB unified memory) for running local AI models. On the $4,300 64GB... og:url: https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb og:image: https://api.daily.dev/og/posts/DZJNmjPfb.png og:image:alt: Running local AI models on Microsoft’s Surface Laptop Ultra: What we found og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Running local AI models on Microsoft’s Surface Laptop Ultra: What we found **[The New Stack](https://daily.dev/sources/newstack)** · 7 min read · 0 upvotes · 0 comments ## Summary Hands-on testing of Microsoft's Surface Laptop Ultra (Nvidia N1X Blackwell GPU, up to 128GB unified memory) for running local AI models. On the $4,300 64GB config, the Nemotron 3.5 Bolt model hit 56-60 tokens/sec, while dense Qwen 3.8 27B at 8-bit only managed 8-9 tokens/sec (improving to 25-28 at 4-bit quantization), and Gemma 4 26B-A4B (MoE) reached ~40 tokens/sec. Performance drops 20-25% on battery, the laptop runs hot and noisy under AI load, and 64GB RAM sits awkwardly between model size tiers. Also covers touchpad haptics, speakers, ports, Windows on ARM maturity, and gaming performance. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How many tokens per second can the Surface Laptop Ultra run Qwen 3.8 27B at 4-bit quantization? The Surface Laptop Ultra runs the dense Qwen 3.8 27B model at 25 to 28 tokens per second when quantized to 4-bit, plugged in. At 8-bit quantization the same model only reaches 8 to 9 tokens per second, a speed limited by memory bandwidth rather than compute, since dense models stress memory bandwidth more than mixture-of-experts models do. _Comparing quantization trade-offs for local inference hardware is easier with benchmarks like these on daily.dev._ ### How much performance does the Surface Laptop Ultra lose when running AI models on battery instead of plugged in? Expect roughly a 20% to 25% performance drop for AI workloads when unplugged, despite Microsoft's claim of retaining 99.6% of performance on battery. The 4-bit Qwen 3.8 27B model dropped from 25-28 tokens per second to about 18 tokens per second, and Nemotron 3.5 Bolt fell from 56-60 to 42-45 tokens per second on battery. _Developers weighing portable AI rigs can track real-world battery benchmarks like this on daily.dev._ ### How fast does the Nemotron 3.5 Bolt model run on the Surface Laptop Ultra's Nvidia N1X Blackwell GPU? Nemotron 3.5 Bolt, a heavily quantized model with 9B active parameters, answered queries at 56 to 60 tokens per second on the Surface Laptop Ultra's N1X Blackwell GPU while plugged in, making it well suited for use in agents like Hermes Agent or OpenCode. _Anyone sizing models against new laptop GPUs can follow benchmarks like this on daily.dev._ ## Similar posts on daily.dev - [I ran local AI models on a six-year-old laptop with no GPU, and they actually worked](https://daily.dev/posts/i-ran-local-ai-models-on-a-six-year-old-laptop-with-no-gpu-and-they-actually-worked-v9yuny0hp) · XDA Developers · 2 upvotes · 0 comments - [Run AI Models Locally: A New Laptop Era Begins](https://daily.dev/posts/run-ai-models-locally-a-new-laptop-era-begins-r1b4a26qf) · Hacker News · 4 upvotes · 0 comments - [Microsoft’s new Surface Laptop has a trackpad that rumbles like a game controller](https://daily.dev/posts/microsoft-s-new-surface-laptop-has-a-trackpad-that-rumbles-like-a-game-controller-bkdujpynk) · The Next Web · 0 upvotes · 0 comments --- Tags: [#data-science](https://daily.dev/tags/data-science), [#local-ai](https://daily.dev/tags/local-ai), [#ollama](https://daily.dev/tags/ollama) [View this post on daily.dev](https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Running local AI models on Microsoft’s Surface Laptop Ultra: What we found","url":"https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb"},"datePublished":"2026-10-16T12:04:59.793Z","dateModified":"2026-10-16T12:05:25.714Z","description":"Hands-on testing of Microsoft's Surface Laptop Ultra (Nvidia N1X Blackwell GPU, up to 128GB unified memory) for running local AI models. On the $4,300 64GB...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/487806477ee7158595e53d22b7d21f40?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/487806477ee7158595e53d22b7d21f40?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"The New Stack","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"The New Stack","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/newstack","url":"https://daily.dev/sources/newstack"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"data-science,local-ai,ollama","timeRequired":"PT7M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"The New Stack","item":"https://daily.dev/sources/newstack"},{"@type":"ListItem","position":3,"name":"Running local AI models on Microsoft’s Surface Laptop Ultra: What we found"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/running-local-ai-models-on-microsoft-s-surface-laptop-ultra-what-we-found-dzjnmjpfb#faq","mainEntity":[{"@type":"Question","name":"How many tokens per second can the Surface Laptop Ultra run Qwen 3.8 27B at 4-bit quantization?","acceptedAnswer":{"@type":"Answer","text":"The Surface Laptop Ultra runs the dense Qwen 3.8 27B model at 25 to 28 tokens per second when quantized to 4-bit, plugged in. At 8-bit quantization the same model only reaches 8 to 9 tokens per second, a speed limited by memory bandwidth rather than compute, since dense models stress memory bandwidth more than mixture-of-experts models do. Comparing quantization trade-offs for local inference hardware is easier with benchmarks like these on daily.dev."}},{"@type":"Question","name":"How much performance does the Surface Laptop Ultra lose when running AI models on battery instead of plugged in?","acceptedAnswer":{"@type":"Answer","text":"Expect roughly a 20% to 25% performance drop for AI workloads when unplugged, despite Microsoft's claim of retaining 99.6% of performance on battery. The 4-bit Qwen 3.8 27B model dropped from 25-28 tokens per second to about 18 tokens per second, and Nemotron 3.5 Bolt fell from 56-60 to 42-45 tokens per second on battery. Developers weighing portable AI rigs can track real-world battery benchmarks like this on daily.dev."}},{"@type":"Question","name":"How fast does the Nemotron 3.5 Bolt model run on the Surface Laptop Ultra's Nvidia N1X Blackwell GPU?","acceptedAnswer":{"@type":"Answer","text":"Nemotron 3.5 Bolt, a heavily quantized model with 9B active parameters, answered queries at 56 to 60 tokens per second on the Surface Laptop Ultra's N1X Blackwell GPU while plugged in, making it well suited for use in agents like Hermes Agent or OpenCode. Anyone sizing models against new laptop GPUs can follow benchmarks like this on daily.dev."}}]} ```