--- title: AI On Your Gaming PC | daily.dev description: Strata, a project by developer Niko1221, lets a 125-billion-parameter LLM run on gaming-grade hardware rather than requiring a datacenter GPU farm. It uses... canonical: https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: AI On Your Gaming PC | daily.dev og:description: Strata, a project by developer Niko1221, lets a 125-billion-parameter LLM run on gaming-grade hardware rather than requiring a datacenter GPU farm. It uses... og:url: https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7 og:image: https://api.daily.dev/og/posts/A7iaDECz7.png og:image:alt: AI On Your Gaming PC 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. # AI On Your Gaming PC **[Hackaday](https://daily.dev/sources/hackaday)** · 3 min read · 2 upvotes · 2 comments ## Summary Strata, a project by developer Niko1221, lets a 125-billion-parameter LLM run on gaming-grade hardware rather than requiring a datacenter GPU farm. It uses Qwen3.8 model variants, including a mixture-of-experts version with 24,576 small experts where only ten are active per token. VRAM acts as a cache for frequently used experts, the bulk of the model stays in system RAM, and a roughly 29 GB lookup table lives on the SSD. Speculative decoding via multi-token prediction boosts throughput, with an RTX 5070 (12 GB VRAM) reportedly hitting 50-90 tokens per second depending on quantization. It exposes OpenAI- and Anthropic-compatible local APIs, though answer quality trails larger hosted models, and realistic requirements are around 32 GB RAM, 12 GB VRAM, and 80 GB storage. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How can I run a 125 billion parameter LLM on a gaming PC instead of a GPU server farm? A project called Strata enables this by treating VRAM as a cache for a much larger model, keeping frequently used mixture-of-experts weights on the GPU while the full set stays in system RAM and a roughly 29 GB lookup table remains on the SSD. It uses Qwen3.8 model variants, including a mixture-of-experts version with 24,576 small experts of which only ten activate per token. _daily.dev surfaces projects like this for developers tracking local LLM inference techniques._ ### What token generation speed can I expect running Qwen3.8-Flash-Next locally with an RTX 5070? An RTX 5070 with 12 GB of VRAM can produce roughly 50 to 90 tokens per second running Qwen3.8-Flash-Next through Strata, depending on the quantization used. This is achieved partly through speculative decoding, which uses the model's multi-token prediction to check several candidate tokens in a single pass. _Developers benchmarking local LLM setups can follow performance reports like this on daily.dev._ ### What hardware do I need to run Strata for local LLM inference? Strata realistically requires at least 32 GB of system RAM, 12 GB of VRAM, and around 80 GB of storage, despite being marketed as running on modest gaming hardware. It also needs OpenAI- and Anthropic-compatible API support on localhost, letting existing chat front ends and coding assistants connect without special handling. _Those planning local LLM hardware budgets can track setups like this through daily.dev._ ## Community discussion Top comments from developers on daily.dev. **@akkitto** · 0 upvotes > Nice! Will try it out. **@akkitto** · 0 upvotes > Looks good, but I quickly hit model limitations. I got a higher model and it went pretty quick from my high-end NVMe, but the model was stuck in a reasoning loop, constantly looping over a single question it asked itself, not stopping, ever... ## Similar posts on daily.dev - [Local AI is more accessible than ever, but with one major GPU-sized caveat](https://daily.dev/posts/local-ai-is-more-accessible-than-ever-but-with-one-major-gpu-sized-caveat-ozobd3gds) · XDA Developers · 0 upvotes · 0 comments - [High-VRAM GPUs aren't the future of local AI — unified memory and Mixture of Experts models are](https://daily.dev/posts/high-vram-gpus-aren-t-the-future-of-local-ai-unified-memory-and-mixture-of-experts-models-are-jdokpgomz) · XDA Developers · 8 upvotes · 0 comments - [Running AI Locally in 2026: What the Hardware Actually Costs](https://daily.dev/posts/running-ai-locally-in-2026-what-the-hardware-actually-costs-zdwnwjdql) · Medium · 0 upvotes · 0 comments --- Tags: [#gpu](https://daily.dev/tags/gpu), [#local-ai](https://daily.dev/tags/local-ai), [#mixture-of-experts](https://daily.dev/tags/mixture-of-experts) [View this post on daily.dev](https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7) ```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":"AI On Your Gaming PC","url":"https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7"},"datePublished":"2026-10-04T11:06:06.442Z","dateModified":"2026-10-04T11:06:27.349Z","description":"Strata, a project by developer Niko1221, lets a 125-billion-parameter LLM run on gaming-grade hardware rather than requiring a datacenter GPU farm. It uses...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/38d611befd5682d34dc5bd199e528596?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/38d611befd5682d34dc5bd199e528596?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Hackaday","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":"Hackaday","logo":"https://media.daily.dev/image/upload/s--JhDfhy70--/f_auto,q_auto/v1767541814/logos/hackaday","url":"https://daily.dev/sources/hackaday"},"commentCount":2,"discussionUrl":"https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":2}],"keywords":"gpu,local-ai,mixture-of-experts","timeRequired":"PT3M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hackaday","item":"https://daily.dev/sources/hackaday"},{"@type":"ListItem","position":3,"name":"AI On Your Gaming PC"}]} {"@context":"https://schema.org","@type":"WebPage","@id":"https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7","comment":[{"@type":"Comment","text":"Nice! Will try it out.","datePublished":"2026-10-04T14:21:53.671Z","url":"https://daily.dev/posts/A7iaDECz7#c-xdTzjHvP2","author":{"@type":"Person","name":"Daniel","url":"https://daily.dev/akkitto","image":"https://media.daily.dev/image/upload/s--FtwJqX4c--/f_auto/v1754900041/avatars/avatar_29TCpY2hJR72V3BlxPXzX?_a=BAMClqZW0"}},{"@type":"Comment","text":"Looks good, but I quickly hit model limitations. I got a higher model and it went pretty quick from my high-end NVMe, but the model was stuck in a reasoning loop, constantly looping over a single question it asked itself, not stopping, ever…","datePublished":"2026-10-04T21:38:36.578Z","url":"https://daily.dev/posts/A7iaDECz7#c-vdOF6P9gT","author":{"@type":"Person","name":"Daniel","url":"https://daily.dev/akkitto","image":"https://media.daily.dev/image/upload/s--FtwJqX4c--/f_auto/v1754900041/avatars/avatar_29TCpY2hJR72V3BlxPXzX?_a=BAMClqZW0"}}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/ai-on-your-gaming-pc-a7iadecz7#faq","mainEntity":[{"@type":"Question","name":"How can I run a 125 billion parameter LLM on a gaming PC instead of a GPU server farm?","acceptedAnswer":{"@type":"Answer","text":"A project called Strata enables this by treating VRAM as a cache for a much larger model, keeping frequently used mixture-of-experts weights on the GPU while the full set stays in system RAM and a roughly 29 GB lookup table remains on the SSD. It uses Qwen3.8 model variants, including a mixture-of-experts version with 24,576 small experts of which only ten activate per token. daily.dev surfaces projects like this for developers tracking local LLM inference techniques."}},{"@type":"Question","name":"What token generation speed can I expect running Qwen3.8-Flash-Next locally with an RTX 5070?","acceptedAnswer":{"@type":"Answer","text":"An RTX 5070 with 12 GB of VRAM can produce roughly 50 to 90 tokens per second running Qwen3.8-Flash-Next through Strata, depending on the quantization used. This is achieved partly through speculative decoding, which uses the model's multi-token prediction to check several candidate tokens in a single pass. Developers benchmarking local LLM setups can follow performance reports like this on daily.dev."}},{"@type":"Question","name":"What hardware do I need to run Strata for local LLM inference?","acceptedAnswer":{"@type":"Answer","text":"Strata realistically requires at least 32 GB of system RAM, 12 GB of VRAM, and around 80 GB of storage, despite being marketed as running on modest gaming hardware. It also needs OpenAI- and Anthropic-compatible API support on localhost, letting existing chat front ends and coding assistants connect without special handling. Those planning local LLM hardware budgets can track setups like this through daily.dev."}}]} ```