--- title: New model scheduling · Ollama Blog | daily.dev description: Ollama shipped a new model scheduling system that computes exact memory requirements instead of estimating, cutting out-of-memory crashes, improving GPU... canonical: https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: New model scheduling · Ollama Blog | daily.dev og:description: Ollama shipped a new model scheduling system that computes exact memory requirements instead of estimating, cutting out-of-memory crashes, improving GPU... og:url: https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7 og:image: https://api.daily.dev/og/posts/DlgSwkSh7.png og:image:alt: New model scheduling · Ollama Blog 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. # New model scheduling · Ollama Blog **[Ollama](https://daily.dev/sources/ollama)** · 2 min read · 0 upvotes · 0 comments ## Summary Ollama shipped a new model scheduling system that computes exact memory requirements instead of estimating, cutting out-of-memory crashes, improving GPU utilization, and boosting multi-GPU performance. Tools like nvidia-smi now match ollama ps for accurate memory reporting. Benchmarks show notable gains: on an RTX 4090 running gemma3:12b with 128k context, token generation jumped from 52 to 85.5 tokens/s with all layers now loaded on GPU. Models on Ollama's new engine (gpt-oss, llama4, gemma3, qwen3, mistral-small3.2, embedding models, etc.) get this by default, with more migrating soon. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### Why did Ollama's GPU memory reporting not match nvidia-smi output? Older versions of Ollama estimated memory requirements rather than measuring them exactly, causing mismatches between ollama ps and tools like nvidia-smi. The new scheduling system introduced in Ollama measures exact memory needed ahead of running a model, so reported usage now matches nvidia-smi, and over-allocation-related out-of-memory crashes are significantly reduced. _daily.dev surfaces updates like this for developers troubleshooting GPU memory mismatches in local LLM setups._ ### How much faster is Ollama after the new model scheduling update on multi-GPU setups? Running mistral-small3.2 on two RTX 4090 GPUs with 32k context, prompt evaluation speed jumped from 127.84 to 1380.24 tokens/s and token generation rose from 43.15 to 55.61 tokens/s after the new scheduling system, which now fits all 41 layers plus the vision model onto the GPUs instead of 40/41. _Developers comparing local LLM throughput across GPU setups can track benchmarks like these on daily.dev._ ### Which Ollama models support the new exact memory scheduling system? Models running on Ollama's new engine get exact memory scheduling by default, including gpt-oss, llama4, gemma3, gemma3n, embeddinggemma, qwen3, qwen3-coder, mistral-small3.2, and all-minilm plus other embedding models. Llama3.2-vision, llama3.2, llama3.1, llama3, and qwen2.5vl are noted as coming soon once they transition to the new engine. _daily.dev helps developers migrating local LLM workloads stay on top of which models gain new engine features._ ## Similar posts on daily.dev - [My 7-year-old GPU runs local AI perfectly, and I don't need my cloud subscriptions anymore](https://daily.dev/posts/my-7-year-old-gpu-runs-local-ai-perfectly-and-i-don-t-need-my-cloud-subscriptions-anymore-lvb6vhriu) · XDA Developers · 2 upvotes · 0 comments - [FreeToken Just Shipped. Is It Better Than Ollama and llama.cpp?](https://daily.dev/posts/freetoken-just-shipped-is-it-better-than-ollama-and-llama-cpp--dsvmi8hqy) · Medium · 0 upvotes · 0 comments - [Untitled](https://daily.dev/posts/untitled-vsbxtb4by) · SitePoint · 0 upvotes · 0 comments - [New coding models & integrations · Ollama Blog](https://daily.dev/posts/new-coding-models-integrations-ollama-blog-onqkepkh1) · Hacker News · 30 upvotes · 0 comments --- Tags: [#gpu](https://daily.dev/tags/gpu), [#local-ai](https://daily.dev/tags/local-ai), [#ai-inference](https://daily.dev/tags/ai-inference), [#ollama](https://daily.dev/tags/ollama) [View this post on daily.dev](https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7) ```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":"New model scheduling · Ollama Blog","url":"https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7"},"datePublished":"2026-08-23T12:20:49.644Z","dateModified":"2026-08-23T12:42:18.941Z","description":"Ollama shipped a new model scheduling system that computes exact memory requirements instead of estimating, cutting out-of-memory crashes, improving GPU...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/03fb1a10ef158f8986c9778334c2af42?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/03fb1a10ef158f8986c9778334c2af42?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Ollama","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":"Ollama","logo":"https://media.daily.dev/image/upload/s--dRYYkBCj--/f_auto,q_auto/v1787487632/logos/ollama-blog?_a=BAMAMicg0","url":"https://daily.dev/sources/ollama"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"gpu,local-ai,ai-inference,ollama","timeRequired":"PT2M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Ollama","item":"https://daily.dev/sources/ollama"},{"@type":"ListItem","position":3,"name":"New model scheduling · Ollama Blog"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/new-model-scheduling-ollama-blog-dlgswksh7#faq","mainEntity":[{"@type":"Question","name":"Why did Ollama's GPU memory reporting not match nvidia-smi output?","acceptedAnswer":{"@type":"Answer","text":"Older versions of Ollama estimated memory requirements rather than measuring them exactly, causing mismatches between ollama ps and tools like nvidia-smi. The new scheduling system introduced in Ollama measures exact memory needed ahead of running a model, so reported usage now matches nvidia-smi, and over-allocation-related out-of-memory crashes are significantly reduced. daily.dev surfaces updates like this for developers troubleshooting GPU memory mismatches in local LLM setups."}},{"@type":"Question","name":"How much faster is Ollama after the new model scheduling update on multi-GPU setups?","acceptedAnswer":{"@type":"Answer","text":"Running mistral-small3.2 on two RTX 4090 GPUs with 32k context, prompt evaluation speed jumped from 127.84 to 1380.24 tokens/s and token generation rose from 43.15 to 55.61 tokens/s after the new scheduling system, which now fits all 41 layers plus the vision model onto the GPUs instead of 40/41. Developers comparing local LLM throughput across GPU setups can track benchmarks like these on daily.dev."}},{"@type":"Question","name":"Which Ollama models support the new exact memory scheduling system?","acceptedAnswer":{"@type":"Answer","text":"Models running on Ollama's new engine get exact memory scheduling by default, including gpt-oss, llama4, gemma3, gemma3n, embeddinggemma, qwen3, qwen3-coder, mistral-small3.2, and all-minilm plus other embedding models. Llama3.2-vision, llama3.2, llama3.1, llama3, and qwen2.5vl are noted as coming soon once they transition to the new engine. daily.dev helps developers migrating local LLM workloads stay on top of which models gain new engine features."}}]} ```