{"id":790,"date":"2026-06-06T01:59:14","date_gmt":"2026-06-06T01:59:14","guid":{"rendered":"https:\/\/convly.ai\/ollama-system-requirements-2026\/"},"modified":"2026-06-15T18:18:09","modified_gmt":"2026-06-15T18:18:09","slug":"ollama-system-requirements-2026","status":"publish","type":"post","link":"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/","title":{"rendered":"Ollama System Requirements in 2026: How Much RAM and VRAM You Really Need"},"content":{"rendered":"<p>The single most common reason a model won&#8217;t run in Ollama isn&#8217;t a bug \u2014 it&#8217;s that the model is bigger than your memory. Ollama itself is tiny; the models are what demand hardware. This guide gives you the real RAM and VRAM numbers for each model size in 2026, plus a simple formula so you know what fits <em>before<\/em> you spend ten minutes downloading something that won&#8217;t load.<\/p>\n<p>If you haven&#8217;t installed Ollama yet, start with our <a href=\"https:\/\/convly.ai\/de\/how-to-install-ollama-2026\/\">Schritt-f\u00fcr-Schritt-Anleitung zur Installation<\/a>.<\/p>\n<div class=\"convly-tldr\">\n<h3>Wichtigste Erkenntnisse<\/h3>\n<ul>\n<li><strong>The rule of thumb:<\/strong> a quantized (Q4) model needs roughly <strong>0.6 GB of memory per billion parameters<\/strong>, plus headroom for context.<\/li>\n<li><strong>2\u20133B models:<\/strong> run on CPU, ~2\u20134 GB RAM. Fine on a basic laptop.<\/li>\n<li><strong>7\u20138B models:<\/strong> ~6\u20138 GB RAM\/VRAM. The sweet spot for most laptops.<\/li>\n<li><strong>27\u201334B models:<\/strong> ~20\u201324 GB VRAM. Needs a high-end GPU or Apple Silicon with lots of unified memory.<\/li>\n<li><strong>70B+ models:<\/strong> 40 GB+ \u2014 a workstation GPU, multi-GPU rig, or 64 GB+ unified memory.<\/li>\n<\/ul>\n<\/div>\n<div id=\"ez-toc-container\" class=\"ez-toc-v2_0_84 counter-flat ez-toc-counter ez-toc-container-direction\">\n<label for=\"ez-toc-cssicon-toggle-item-6a38aed776184\" class=\"ez-toc-cssicon-toggle-label\"><span class=\"\"><span class=\"eztoc-hide\" style=\"display:none;\">Umschalten<\/span><span class=\"ez-toc-icon-toggle-span\"><svg style=\"fill: #000000;color:#000000\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" class=\"list-377408\" width=\"20px\" height=\"20px\" viewbox=\"0 0 24 24\" fill=\"none\"><path d=\"M6 6H4v2h2V6zm14 0H8v2h12V6zM4 11h2v2H4v-2zm16 0H8v2h12v-2zM4 16h2v2H4v-2zm16 0H8v2h12v-2z\" fill=\"currentColor\"><\/path><\/svg><svg style=\"fill: #000000;color:#000000\" class=\"arrow-unsorted-368013\" xmlns=\"http:\/\/www.w3.org\/2000\/svg\" width=\"10px\" height=\"10px\" viewbox=\"0 0 24 24\" version=\"1.2\" baseprofile=\"tiny\"><path d=\"M18.2 9.3l-6.2-6.3-6.2 6.3c-.2.2-.3.4-.3.7s.1.5.3.7c.2.2.4.3.7.3h11c.3 0 .5-.1.7-.3.2-.2.3-.5.3-.7s-.1-.5-.3-.7zM5.8 14.7l6.2 6.3 6.2-6.3c.2-.2.3-.5.3-.7s-.1-.5-.3-.7c-.2-.2-.4-.3-.7-.3h-11c-.3 0-.5.1-.7.3-.2.2-.3.5-.3.7s.1.5.3.7z\"\/><\/svg><\/span><\/span><\/label><input type=\"checkbox\"  id=\"ez-toc-cssicon-toggle-item-6a38aed776184\"  aria-label=\"Umschalten\" \/><nav><ul class='ez-toc-list ez-toc-list-level-1' ><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-1\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#Why_memory_is_the_whole_story\" >Why memory is the whole story<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-2\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#The_simple_formula\" >The simple formula<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-3\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#Requirements_by_model_size\" >Requirements by model size<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-4\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#GPU_vs_CPU_vs_Apple_Silicon\" >GPU vs CPU vs Apple Silicon<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-5\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#How_to_make_a_big_model_fit\" >How to make a big model fit<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-6\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#Storage_and_software_prerequisites_people_forget\" >Storage and software prerequisites people forget<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-7\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#FAQ\" >H\u00e4ufig gestellte Fragen (FAQ)<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-8\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#Bottom_line\" >Fazit<\/a><\/li><li class='ez-toc-page-1'><a class=\"ez-toc-link ez-toc-heading-9\" href=\"https:\/\/convly.ai\/de\/ollama-system-requirements-2026\/#Related_articles\" >Verwandte Artikel<\/a><\/li><\/ul><\/nav><\/div>\n<h2><span class=\"ez-toc-section\" id=\"Why_memory_is_the_whole_story\"><\/span>Why memory is the whole story<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>To generate text, a model&#8217;s weights have to sit in fast memory \u2014 your GPU&#8217;s VRAM, or system RAM if you&#8217;re running on CPU. If the model doesn&#8217;t fit, one of two things happens: Ollama spills part of it to slower memory (and performance collapses), or it refuses to load with an out-of-memory error. Everything else \u2014 CPU speed, disk, OS \u2014 matters far less than having enough of the right memory.<\/p>\n<p>Two factors set the requirement:<\/p>\n<ol>\n<li><strong>Parameter count<\/strong> \u2014 a 7B model has 7 billion weights; a 70B model has ten times as many.<\/li>\n<li><strong>Quantisierung<\/strong> \u2014 Ollama uses compressed GGUF weights. A 4-bit (Q4) quant cuts memory roughly in half versus 8-bit, with minimal quality loss, which is why it&#8217;s the default sweet spot.<\/li>\n<\/ol>\n<h2><span class=\"ez-toc-section\" id=\"The_simple_formula\"><\/span>The simple formula<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>For a 4-bit quantized model \u2014 what Ollama pulls by default \u2014 estimate:<\/p>\n<blockquote>\n<p><strong>Memory needed \u2248 (parameters in billions) \u00d7 0.6 GB + context overhead<\/strong><\/p>\n<\/blockquote>\n<p>So a 7B model needs roughly 4\u20135 GB, a 13B model about 8 GB, a 27B model around 18\u201320 GB, and a 70B model 40 GB or more. Add a bit on top for the KV cache, which grows with how long your conversations get. Always leave a few gigabytes of headroom for your operating system.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Requirements_by_model_size\"><\/span>Requirements by model size<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<table class=\"convly-vs\">\n<thead>\n<tr>\n<th>Modellgr\u00f6\u00dfe<\/th>\n<th>Memory (Q4)<\/th>\n<th>Runs on<\/th>\n<th>Example models<\/th>\n<\/tr>\n<\/thead>\n<tbody>\n<tr>\n<td>2\u20133B<\/td>\n<td>~2\u20134 GB<\/td>\n<td>CPU \/ any laptop<\/td>\n<td>Gemma2 2B, Phi-4 mini<\/td>\n<\/tr>\n<tr>\n<td>7\u20138B<\/td>\n<td>~6\u20138 GB<\/td>\n<td>Entry GPU \/ 16 GB laptop<\/td>\n<td>DeepSeek-R1 7B, Llama 3.3 8B<\/td>\n<\/tr>\n<tr>\n<td>13\u201314B<\/td>\n<td>~10\u201312 GB<\/td>\n<td>Mid-range GPU<\/td>\n<td>Phi-4, mid Qwen<\/td>\n<\/tr>\n<tr>\n<td>27\u201334B<\/td>\n<td>~18\u201324 GB<\/td>\n<td>High-end GPU \/ Apple Silicon<\/td>\n<td>Gemma 4 26B, Qwen 3.6 27B<\/td>\n<\/tr>\n<tr>\n<td>70B<\/td>\n<td>~40\u201348 GB<\/td>\n<td>Workstation \/ multi-GPU<\/td>\n<td>Llama 70B class<\/td>\n<\/tr>\n<tr>\n<td>200B+ (MoE)<\/td>\n<td>100 GB+<\/td>\n<td>Server \/ huge unified memory<\/td>\n<td>Qwen3 235B-A22B<\/td>\n<\/tr>\n<\/tbody>\n<\/table>\n<p>For a deeper breakdown across specific models, see our guide to <a href=\"https:\/\/convly.ai\/de\/vram-requirements-every-major-llm-2026\/\">VRAM-Anforderungen f\u00fcr alle wichtigen Gro\u00dfsprachmodelle (LLMs)<\/a>.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"GPU_vs_CPU_vs_Apple_Silicon\"><\/span>GPU vs CPU vs Apple Silicon<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p><strong>NVIDIA GPU<\/strong> \u2014 the gold standard. VRAM is the hard limit: the model must fit in your card&#8217;s memory to run fast. A 24 GB card (RTX 4090\/5090) comfortably runs up to ~27\u201334B models.<\/p>\n<p><strong>CPU only<\/strong> \u2014 works for small models (2\u20138B) but is much slower, since system RAM bandwidth can&#8217;t match a GPU. Perfectly fine for light tasks on a laptop with no discrete GPU.<\/p>\n<p><strong>Apple Silicon<\/strong> \u2014 a special case, and a strong one. Because Macs use <em>vereinbarter Speicher<\/em> shared between CPU and GPU, a Mac with 64 GB can load models that would need an expensive multi-GPU PC. Since Ollama v0.19 (March 2026) added the MLX backend, Apple Silicon also got much faster \u2014 making a high-memory Mac one of the best single-box local-LLM machines you can buy. For how that stacks up against a discrete GPU, see <a href=\"https:\/\/convly.ai\/de\/amd-strix-halo-vs-apple-m4-pro\/\">Strix Halo vs Apple M4 Pro<\/a>.<\/p>\n<p><strong>AMD GPU<\/strong> \u2014 supported via ROCm. It works well for inference in 2026; check our <a href=\"https:\/\/convly.ai\/de\/amd-rocm-vs-nvidia-cuda-2026\/\">ROCm-gegen-CUDA-Analyse<\/a> for the current state.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"How_to_make_a_big_model_fit\"><\/span>How to make a big model fit<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>If the model you want is just over your memory, you have options before giving up:<\/p>\n<ul>\n<li><strong>Use a smaller quant<\/strong> \u2014 pull a <code>q4<\/code> or even <code>q3<\/code> variant instead of <code>q8<\/code>. You trade a little quality for a big memory saving.<\/li>\n<li><strong>Pick a smaller model size<\/strong> \u2014 a well-chosen 8B often beats a barely-running, swapped-out 27B.<\/li>\n<li><strong>Shorten the context window<\/strong> \u2014 a smaller context uses less KV-cache memory.<\/li>\n<li><strong>Close other apps<\/strong> \u2014 on a CPU\/unified-memory machine, free RAM is your budget.<\/li>\n<\/ul>\n<p>To pick a model matched to your hardware, see the <a href=\"https:\/\/convly.ai\/de\/best-local-llms-to-run-on-ollama-2026\/\">besten lokalen Sprachmodelle f\u00fcr Ollama<\/a>.<\/p>\n<p><!--ai-enriched--><\/p>\n<h2><span class=\"ez-toc-section\" id=\"Storage_and_software_prerequisites_people_forget\"><\/span>Storage and software prerequisites people forget<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>RAM and VRAM get all the attention, but two quieter requirements trip up more first-time installs than anything else: disk space and the software stack underneath. Get these wrong and Ollama either refuses to install or fails halfway through a model download.<\/p>\n<p><strong>Disk space.<\/strong> The Ollama binary itself is small \u2014 budget roughly <strong>4&nbsp;GB<\/strong> for the install. The models are what eat your drive. Every model is downloaded once and cached on disk, then loaded into memory at runtime, so you need room for the full weights on top of whatever you have free. As a rough guide at common 4-bit quantization:<\/p>\n<ul>\n<li><strong>An 8B model<\/strong> (e.g. Llama 3.1 8B): about 5&nbsp;GB on disk.<\/li>\n<li><strong>A 20B-class model:<\/strong> roughly 12\u201314&nbsp;GB.<\/li>\n<li><strong>A 70B model:<\/strong> around 40&nbsp;GB.<\/li>\n<li><strong>A very large MoE model<\/strong> (Llama&nbsp;4-class): 65&nbsp;GB or more.<\/li>\n<\/ul>\n<p>These stack up fast. A casual collection of a few models lands at 30\u201380&nbsp;GB; keep several large variants and you will cross 200&nbsp;GB without trying. A 512&nbsp;GB SSD is a sensible floor if you plan to collect models.<\/p>\n<p><strong>Use an SSD, ideally NVMe.<\/strong> Because the weights are read off disk into RAM or VRAM every time a model first loads, a slow mechanical drive shows up directly as sluggish startup \u2014 a 40&nbsp;GB model crawls off a spinning disk. Fast storage does not change tokens-per-second once the model is loaded, but it makes the first prompt feel instant instead of a 30-second stall.<\/p>\n<p><strong>Operating system and drivers.<\/strong> Ollama runs natively on all three platforms, but each has a floor:<\/p>\n<ul>\n<li><strong>macOS:<\/strong> 11 (Big Sur) or newer, on both Apple Silicon and Intel.<\/li>\n<li><strong>Windows:<\/strong> Windows 10 22H2 or newer (Home or Pro), on x86_64 and ARM64 \u2014 so Snapdragon machines run it natively, without x86 emulation.<\/li>\n<li><strong>Linux:<\/strong> most modern distributions (Ubuntu 18.04+, Debian, Fedora, RHEL, Arch).<\/li>\n<\/ul>\n<p>For GPU acceleration you also need current drivers: a recent NVIDIA driver \u2014 <strong>531 or newer<\/strong> (and 570 or newer for older Maxwell- and Pascal-era cards) \u2014 for CUDA, or a Vulkan-capable or ROCm v7 driver stack on AMD Radeon. Miss the driver and Ollama silently falls back to CPU \u2014 which is the most common reason a machine &#8220;with a good GPU&#8221; runs slowly.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"FAQ\"><\/span>H\u00e4ufig gestellte Fragen (FAQ)<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<h3>How much RAM do I need to run Ollama?<\/h3>\n<p>It depends entirely on the model. Ollama itself needs almost nothing; the model sets the requirement. As a rule, a 4-bit model needs about 0.6 GB per billion parameters \u2014 so ~4\u20135 GB for a 7B model, ~8 GB for 13B, and 40 GB+ for a 70B. Always leave a few gigabytes free for your OS.<\/p>\n<h3>Kann ich Ollama ohne GPU ausf\u00fchren?<\/h3>\n<p>Yes. Small models (2\u20138B) run fine on CPU, just more slowly than on a GPU. A model like Gemma2 2B needs only about 1.7 GB of RAM and works on basic laptops. For models above ~13B, a GPU or Apple Silicon with unified memory makes a real difference.<\/p>\n<h3>How much VRAM do I need for a 7B model?<\/h3>\n<p>About 6\u20138 GB for a 4-bit quantized 7B model, including some context overhead. That fits comfortably on most entry-level discrete GPUs and on laptops with 16 GB of unified or system memory.<\/p>\n<h3>Why is Ollama running so slowly?<\/h3>\n<p>Almost always because the model doesn&#8217;t fully fit in your GPU&#8217;s VRAM, so part of it spilled to system RAM or CPU. Check with <code>ollama ps<\/code> \u2014 if it shows high CPU usage, switch to a smaller model or a more aggressive quant so the whole model fits in fast memory.<\/p>\n<h3>Is a Mac good for running Ollama?<\/h3>\n<p>Yes, often excellent. Apple Silicon&#8217;s unified <a href=\"https:\/\/convly.ai\/de\/llm-vram-calculator\/\"  data-wpil-monitor-id=\"62\">memory lets a 64 GB Mac run models<\/a> that would otherwise need a costly multi-GPU PC, and the MLX backend (since v0.19) made it fast too. A high-memory Mac is one of the best single-machine options for local LLMs in 2026.<\/p>\n<h3>How much disk space do I need for Ollama?<\/h3>\n<p>Plan for about 4&nbsp;GB for the Ollama install itself, then add the size of each model you pull. At 4-bit quantization an 8B model is roughly 5&nbsp;GB, a 70B is around 40&nbsp;GB, and the largest models exceed 65&nbsp;GB. A typical multi-model setup lands between 30 and 80&nbsp;GB, so a 512&nbsp;GB SSD is a comfortable starting point. An SSD (preferably NVMe) is strongly recommended, because models load off disk every time you first run them.<\/p>\n<h3>Where does Ollama store models, and can I move them to another drive?<\/h3>\n<p>By default Ollama keeps downloaded models in a hidden folder in your home directory \u2014 <strong>~\/.ollama<\/strong> on macOS and Linux, and <strong>%HOMEPATH%.ollama<\/strong> on Windows. If your system drive is small, you can redirect storage to a larger or external disk by setting the <strong>OLLAMA_MODELS<\/strong> environment variable to a new path before starting Ollama. This is the cleanest fix when your boot drive runs out of room.<\/p>\n<h3>Which operating systems does Ollama support?<\/h3>\n<p>Ollama runs natively on macOS 11 (Big Sur) or newer, Windows 10 22H2 or newer (64-bit, including ARM64 devices like Snapdragon laptops), and most modern Linux distributions such as Ubuntu 18.04+, Fedora, and Arch. For GPU acceleration you also need an up-to-date driver \u2014 a recent NVIDIA driver for CUDA, or a ROCm\/Vulkan-capable driver on AMD \u2014 otherwise Ollama runs on the CPU instead.<\/p>\n<h2><span class=\"ez-toc-section\" id=\"Bottom_line\"><\/span>Fazit<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<p>Before you download anything, do the quick math: parameters \u00d7 0.6 GB for a 4-bit model, plus headroom. Match that to your VRAM (NVIDIA\/AMD) or unified memory (Apple), and you&#8217;ll never hit a frustrating out-of-memory error again. When in doubt, start one size smaller than you think \u2014 a model that fits and runs fast beats a bigger one that crawls.<\/p>\n<p><!--related-block--><\/p>\n<div class=\"convly-related\">\n<h2><span class=\"ez-toc-section\" id=\"Related_articles\"><\/span>Verwandte Artikel<span class=\"ez-toc-section-end\"><\/span><\/h2>\n<ul>\n<li><a href=\"https:\/\/convly.ai\/de\/ollama-vs-jan-2026\/\">Ollama vs. Jan: Welche lokale KI-Anwendung gewinnt 2026?<\/a><\/li>\n<li><a href=\"https:\/\/convly.ai\/de\/lm-studio-complete-guide-2026\/\">LM Studio: Der umfassende Leitfaden (2026)<\/a><\/li>\n<li><a href=\"https:\/\/convly.ai\/de\/claude-5-new-ai-models-june-2026\/\">Gibt es einen Claude 5? Claude Fable 5 und alle wichtigen KI-Modelle im Juni 2026<\/a><\/li>\n<li><a href=\"https:\/\/convly.ai\/de\/llm-hallucinations-complete-guide\/\">LLM-Halluzinationen im Jahr 2026: Warum sie auftreten und wie man sie verhindert<\/a><\/li>\n<li><a href=\"https:\/\/convly.ai\/de\/prompt-engineering-techniques\/\">Prompt-Engineering im Jahr 2026: 12 Techniken, die tats\u00e4chlich funktionieren<\/a><\/li>\n<li><a href=\"https:\/\/convly.ai\/de\/what-is-ollama-complete-guide-2026\/\">Was ist Ollama? Der umfassende Leitfaden zum lokalen Betrieb von LLMs im Jahr 2026<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>The number-one reason a model fails to run isn&#8217;t a bug \u2014 it&#8217;s memory. Here&#8217;s exactly how much RAM and VRAM each Ollama model size needs, and a formula to know before you download.<\/p>","protected":false},"author":1,"featured_media":796,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"site-sidebar-layout":"default","site-content-layout":"","ast-site-content-layout":"default","site-content-style":"default","site-sidebar-style":"default","ast-global-header-display":"","ast-banner-title-visibility":"","ast-main-header-display":"","ast-hfb-above-header-display":"","ast-hfb-below-header-display":"","ast-hfb-mobile-header-display":"","site-post-title":"","ast-breadcrumbs-content":"","ast-featured-img":"","footer-sml-layout":"","ast-disable-related-posts":"","theme-transparent-header-meta":"","adv-header-id-meta":"","stick-header-meta":"","header-above-stick-meta":"","header-main-stick-meta":"","header-below-stick-meta":"","astra-migrate-meta-layouts":"default","ast-page-background-enabled":"default","ast-page-background-meta":{"desktop":{"background-color":"var(--ast-global-color-5)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"ast-content-background-meta":{"desktop":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"tablet":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""},"mobile":{"background-color":"var(--ast-global-color-4)","background-image":"","background-repeat":"repeat","background-position":"center center","background-size":"auto","background-attachment":"scroll","background-type":"","background-media":"","overlay-type":"","overlay-color":"","overlay-opacity":"","overlay-gradient":""}},"footnotes":""},"categories":[3],"tags":[642,643,640,644,641,639],"class_list":["post-790","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-llms","tag-local-llm-vram","tag-ollama-gpu","tag-ollama-hardware-requirements","tag-ollama-ram","tag-ollama-requirements","tag-ollama-system-requirements"],"_links":{"self":[{"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/posts\/790","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/comments?post=790"}],"version-history":[{"count":5,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/posts\/790\/revisions"}],"predecessor-version":[{"id":1172,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/posts\/790\/revisions\/1172"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/media\/796"}],"wp:attachment":[{"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/media?parent=790"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/categories?post=790"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/convly.ai\/de\/wp-json\/wp\/v2\/tags?post=790"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}