--- title: Qwen 3.8 27B on Strix Halo: 18 t/s With MTP, Nearly 2x... description: Qwen 3.8 27B achieves 18 tokens/second with MTP (multi-token prediction) speculative decoding on AMD Strix Halo hardware, nearly double the 9.6 t/s baseline.... canonical: https://daily.dev/posts/jodyebgl9 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Qwen 3.8 27B on Strix Halo: 18 t/s With MTP, Nearly 2x the Baseline | daily.dev og:description: Qwen 3.8 27B achieves 18 tokens/second with MTP (multi-token prediction) speculative decoding on AMD Strix Halo hardware, nearly double the 9.6 t/s baseline.... og:url: https://daily.dev/posts/jodyebgl9 og:image: https://api.daily.dev/og/posts/jOdyeBgL9.png og:image:alt: Post cover image 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. # Qwen 3.8 27B on Strix Halo: 18 t/s With MTP, Nearly 2x the Baseline **[Qwen](https://daily.dev/sources/qwen)** · [@geniusatwork](https://daily.dev/geniusatwork) · 2 upvotes · 0 comments ## Summary Qwen 3.8 27B achieves 18 tokens/second with MTP (multi-token prediction) speculative decoding on AMD Strix Halo hardware, nearly double the 9.6 t/s baseline. The winning configuration uses the vulkan-radv toolbox (which beats amdvlk and ROCm, the latter unable to load Qwen 3.8 at all), draft-n-max 2 (which outperforms n-max 5 by 19% due to rejection overhead at Q6_K quantization), and froggeric's fixed chat template v22 to fix official template bugs around thinking mode, tool calling, and reasoning_effort. Enabling preserve_thinking is essential for keeping the KV prefix cache valid across multi-turn conversations. The final result matches the Q8_0 tier of Qwen 3.6 MTP in speed while offering a newer, better-benchmarked model. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Similar posts on daily.dev - [Don’t just attend KubeCon \+ CloudNativeCon, Merge Forward your experience\!](https://daily.dev/posts/don-t-just-attend-kubecon-cloudnativecon-merge-forward-your-experience--l0rpp73x8) · CNCF · 1 upvotes · 0 comments - [Announcing H2 2026 KCDs](https://daily.dev/posts/announcing-h2-2026-kcds-m96goajm1) · CNCF · 1 upvotes · 0 comments - [Two months of Open Community Groups](https://daily.dev/posts/two-months-of-open-community-groups-asf52zhbs) · CNCF · 0 upvotes · 0 comments - [CNCF Unveils Schedule for KubeCon \+ CloudNativeCon Europe 2026](https://daily.dev/posts/cncf-unveils-schedule-for-kubecon-cloudnativecon-europe-2026-ikhcoa5cb) · CNCF · 2 upvotes · 0 comments - [CNCF Debuts KubeCon \+ CloudNativeCon Japan 2026 Schedule](https://daily.dev/posts/cncf-debuts-kubecon-cloudnativecon-japan-2026-schedule-xp5pyudub) · CNCF · 1 upvotes · 0 comments --- Tags: [#vulkan](https://daily.dev/tags/vulkan), [#local-ai](https://daily.dev/tags/local-ai), [#ai-inference](https://daily.dev/tags/ai-inference), [#qwen](https://daily.dev/tags/qwen), [#llama-cpp](https://daily.dev/tags/llama-cpp) [View this post on daily.dev](https://daily.dev/posts/jodyebgl9) ```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":"DiscussionForumPosting","mainEntityOfPage":"https://daily.dev/posts/jodyebgl9","headline":"Qwen 3.8 27B on Strix Halo: 18 t/s With MTP, Nearly 2x the Baseline","text":"Shared: Qwen 3.8 27B on Strix Halo: 18 t/s With MTP, Nearly 2x the Baseline","url":"https://daily.dev/posts/jodyebgl9","datePublished":"2026-08-17T12:56:29.104Z","dateModified":"2026-08-17T12:57:42.906Z","author":{"@type":"Person","name":"Sam Dolin","url":"https://daily.dev/geniusatwork","image":"https://media.daily.dev/image/upload/s--nUhoZxKm--/f_auto/v1787434895/avatars/avatar_75QyPrL3BZgArSo6aFWuZ?_a=BAMAMicg0","interactionStatistic":{"@type":"InteractionCounter","interactionType":{"@type":"EndorseAction"},"userInteractionCount":1020}},"image":"https://media.daily.dev/image/upload/s--ZrL_HSsR--/f_auto/v1722860399/public/Placeholder%2006","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"sharedContent":{"@type":"WebPage","url":"https://api.daily.dev/r/Fi1KPEF9N"},"isPartOf":{"@type":"WebPage","url":"https://daily.dev/squads/qwen","name":"Qwen"}} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Qwen","item":"https://daily.dev/squads/qwen"},{"@type":"ListItem","position":3,"name":"Qwen 3.8 27B on Strix Halo: 18 t/s With MTP, Nearly 2x the Baseline"}]} ```