--- title: My years-old gaming PC became the best AI machine in the... description: Swapping an RTX 2070 Super for an RTX 4070 Ti Super doubled VRAM from 8GB to 16GB on an otherwise unchanged years-old gaming PC (Ryzen 5 3600, 32GB DDR4),... canonical: https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: My years-old gaming PC became the best AI machine in the house, and it only needed one upgrade | daily.dev og:description: Swapping an RTX 2070 Super for an RTX 4070 Ti Super doubled VRAM from 8GB to 16GB on an otherwise unchanged years-old gaming PC (Ryzen 5 3600, 32GB DDR4),... og:url: https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd og:image: https://api.daily.dev/og/posts/l8lIMcjcd.png og:image:alt: My years-old gaming PC became the best AI machine in the house, and it only needed one upgrade 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. # My years-old gaming PC became the best AI machine in the house, and it only needed one upgrade **[XDA Developers](https://daily.dev/sources/xda-developers)** · 5 min read · 0 upvotes · 0 comments ## Summary Swapping an RTX 2070 Super for an RTX 4070 Ti Super doubled VRAM from 8GB to 16GB on an otherwise unchanged years-old gaming PC (Ryzen 5 3600, 32GB DDR4), transforming it into a capable local AI inference machine. With 8GB, only small models like Qwen 3.5 4B and Gemma 4 E4B fit fully in VRAM, running at ~80 tokens/sec; anything larger spilled into system RAM and became unusably slow. With 16GB, GPT-OSS-20B (MXFP4, ~12GB) runs at 123 tokens/sec and Gemma 4 12B fits with room for multimodal input, enabling larger context windows, more reliable document analysis, and coding tasks finished in fewer prompt iterations. The CPU stayed largely idle throughout, underscoring that VRAM capacity, not compute, was the bottleneck. Frontier open models like Qwen 3.8 27B and Devstral Small 2 now require 24GB or more, meaning 16GB is already not the new ceiling. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### How much VRAM do I need to run GPT-OSS-20B locally with MXFP4 quantization? GPT-OSS-20B at its native MXFP4 quantization occupies around 12GB of VRAM. On a 16GB GPU like the RTX 4070 Ti Super, it generates about 123 tokens per second at 16K context, since it activates only 3.6B of its 21B total parameters per token, making it feasible on consumer cards with 16GB or more. _daily.dev surfaces real-world VRAM benchmarks for developers sizing local LLM hardware._ ### How much VRAM does Gemma 4 12B need at Q4_K_M quantization? Gemma 4 12B at Q4_K_M quantization uses about 7.66GB of weights, or roughly 8.4GB total including cache at 32K context. This fits comfortably on a 16GB GPU with room left for image and audio input, running at around 63 tokens per second, but it would spill out of an 8GB card and become extremely slow. _Developers picking local model sizes for their GPU budget can track hardware requirement details on daily.dev._ ### How much VRAM does Devstral Small 2 24B require to run locally? Devstral Small 2 requires at least 24GB of VRAM for quantized setups, and roughly double that (around 48GB) for a full unquantized FP16 configuration. This places it, along with other frontier open models like Qwen 3.8 27B, firmly in the 24GB-and-above VRAM tier rather than the 16GB range. _Anyone weighing GPU upgrades against growing local model sizes can follow these thresholds on daily.dev._ ## Similar posts on daily.dev - [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 - [I almost upgraded my GPU to run larger local LLMs, but this 8B model proved I didn't have to](https://daily.dev/posts/i-almost-upgraded-my-gpu-to-run-larger-local-llms-but-this-8b-model-proved-i-didn-t-have-to-gc24e505l) · XDA Developers · 1 upvotes · 0 comments - [Old Nvidia GPUs with 24GB VRAM are crushing new cards at local AI inference, and here's why](https://daily.dev/posts/old-nvidia-gpus-with-24gb-vram-are-crushing-new-cards-at-local-ai-inference-and-here-s-why-itrku8b2e) · XDA Developers · 1 upvotes · 0 comments - [I ditched cloud AI for these 3 local models, and my 8GB GPU handles them all](https://daily.dev/posts/i-ditched-cloud-ai-for-these-3-local-models-and-my-8gb-gpu-handles-them-all-96ixzo7rg) · XDA Developers · 0 upvotes · 0 comments --- Tags: [#nvidia](https://daily.dev/tags/nvidia), [#gpu](https://daily.dev/tags/gpu), [#local-ai](https://daily.dev/tags/local-ai), [#ollama](https://daily.dev/tags/ollama) [View this post on daily.dev](https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd) ```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":"My years-old gaming PC became the best AI machine in the house, and it only needed one upgrade","url":"https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd"},"datePublished":"2026-09-02T20:33:31.761Z","dateModified":"2026-09-02T22:15:15.293Z","description":"Swapping an RTX 2070 Super for an RTX 4070 Ti Super doubled VRAM from 8GB to 16GB on an otherwise unchanged years-old gaming PC (Ryzen 5 3600, 32GB DDR4),...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/66476a40053700c1b6cbb8bfbd2e84da?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/66476a40053700c1b6cbb8bfbd2e84da?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"XDA Developers","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":"XDA Developers","logo":"https://media.daily.dev/image/upload/s--kCg6yyAP--/f_auto,q_auto/v1774964407/logos/xda-developers?_a=BAMAMiWQ0","url":"https://daily.dev/sources/xda-developers"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"nvidia,gpu,local-ai,ollama","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"XDA Developers","item":"https://daily.dev/sources/xda-developers"},{"@type":"ListItem","position":3,"name":"My years-old gaming PC became the best AI machine in the house, and it only needed one upgrade"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/my-years-old-gaming-pc-became-the-best-ai-machine-in-the-house-and-it-only-needed-one-upgrade-l8limcjcd#faq","mainEntity":[{"@type":"Question","name":"How much VRAM do I need to run GPT-OSS-20B locally with MXFP4 quantization?","acceptedAnswer":{"@type":"Answer","text":"GPT-OSS-20B at its native MXFP4 quantization occupies around 12GB of VRAM. On a 16GB GPU like the RTX 4070 Ti Super, it generates about 123 tokens per second at 16K context, since it activates only 3.6B of its 21B total parameters per token, making it feasible on consumer cards with 16GB or more. daily.dev surfaces real-world VRAM benchmarks for developers sizing local LLM hardware."}},{"@type":"Question","name":"How much VRAM does Gemma 4 12B need at Q4_K_M quantization?","acceptedAnswer":{"@type":"Answer","text":"Gemma 4 12B at Q4_K_M quantization uses about 7.66GB of weights, or roughly 8.4GB total including cache at 32K context. This fits comfortably on a 16GB GPU with room left for image and audio input, running at around 63 tokens per second, but it would spill out of an 8GB card and become extremely slow. Developers picking local model sizes for their GPU budget can track hardware requirement details on daily.dev."}},{"@type":"Question","name":"How much VRAM does Devstral Small 2 24B require to run locally?","acceptedAnswer":{"@type":"Answer","text":"Devstral Small 2 requires at least 24GB of VRAM for quantized setups, and roughly double that (around 48GB) for a full unquantized FP16 configuration. This places it, along with other frontier open models like Qwen 3.8 27B, firmly in the 24GB-and-above VRAM tier rather than the 16GB range. Anyone weighing GPU upgrades against growing local model sizes can follow these thresholds on daily.dev."}}]} ```