# Best mini PC for running local LLMs (2026) | What Can I Run? > Mac mini vs Mac Studio vs Framework Desktop vs DGX Spark for local LLMs: usable memory, list prices, estimated tok/s and what each runs at Q4. Source: https://whatcanirunai.vercel.app/best-mini-pc-for-local-llm/ # Best mini PC for running local LLMs Mac mini, Mac Studio, Framework Desktop (Ryzen AI Max+ 395) and NVIDIA DGX Spark: every compact unified-memory desktop we track, ranked by how much memory a model can actually use, with list prices and estimated speed for six popular open models at Q4\_K\_M. Mini PCs tracked 10 Usable memory range 12–384 GB Price range $899–$22,299 Start with the model you want. For 27–32B models, the cheapest in-stock mini PC is the **Mac mini M6 32GB** ($1,799). For 70B models and gpt-oss-120b, the cheapest in-stock mini PC is the **Framework Desktop (Ryzen AI Max+ 395) 128GB** ($3,449). ## Top picks Cheapest way in ### [Mac mini M6 16GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-16/) Usable memory 12 GB Bandwidth 153 GB/s Run at Q4 350 / 656 Runs 7–8B chat models at Q4\. Biggest at Q4: [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/). **$899** Apple list price, M6 12C CPU/12C GPU, 16GB/256GB (MHQK4LL/A), [source](https://www.apple.com/shop/buy-mac/mac-mini/m6-chip-12-core-cpu-12-core-gpu-16gb-memory-256gb-storage), 2026-11-12 [Amazon](https://www.amazon.com/s?k=Mac%20mini%20M6%2016GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=Mac%20mini%20M6%2016GB) Best for 27–32B models ### [Mac mini M6 32GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-32/) Usable memory 24 GB Bandwidth 170 GB/s Run at Q4 423 / 656 The cheapest in-stock mini PC that fits Qwen3.8-27B at Q4\. Biggest at Q4: [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/). **$1799** Apple list price, M6 12C CPU/12C GPU, 32GB/1TB (MHQV4LL/A), [source](https://www.apple.com/shop/buy-mac/mac-mini/m6-chip-12-core-cpu-12-core-gpu-32gb-memory-1tb-storage), 2026-11-12 [Amazon](https://www.amazon.com/s?k=Mac%20mini%20M6%2032GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=Mac%20mini%20M6%2032GB) Best for 70B models · Best for gpt-oss-120b ### [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/) Usable memory 96 GB Bandwidth 256 GB/s Run at Q4 545 / 656 The cheapest in-stock mini PC that fits Llama 3.3 70B at Q4\. Also the cheapest that fits gpt-oss-120b at Q4\. Biggest at Q4: [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/). **$3449** Framework list price, DIY Edition 128GB (pre-order, ships within 5 business days), [source](https://frame.work/products/desktop-diy-amd-aimax300/comparison), 2026-11-05 [Amazon](https://www.amazon.com/s?k=Framework%20Desktop%20%28Ryzen%20AI%20Max%2B%20395%29%20128GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=Framework%20Desktop%20%28Ryzen%20AI%20Max%2B%20395%29%20128GB) Most memory ### [Mac Studio M5 Ultra 512GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-512/) Usable memory 384 GB Bandwidth 1200 GB/s Run at Q4 610 / 656 The most usable memory and bandwidth of any compact desktop we track. Biggest at Q4: [Qwen3-Coder-480B-A35B-Instruct](https://whatcanirunai.vercel.app/models/qwen3-coder-480b-a35b-instruct/). **$22299** Apple list price, M5 Ultra 36C CPU/80C GPU, 512GB/1TB, [source](https://www.apple.com/shop/buy-mac/mac-studio/m5-ultra-chip-36-core-cpu-80-core-gpu-512gb-memory-1tb-storage), 2026-11-12 [Amazon](https://www.amazon.com/s?k=Mac%20Studio%20M5%20Ultra%20512GB&tag=whatcanirunai-20)[eBay](https://www.ebay.com/sch/i.html?%5Fnkw=Mac%20Studio%20M5%20Ultra%20512GB) We may earn a commission when you buy through these links, at no extra cost to you. ## Every mini PC, by usable memory Usable memory is about 75% of unified memory ([how we calculate](https://whatcanirunai.vercel.app/method/)). Prices are dated list-price snapshots; each row links its source. The DGX Spark is reported sold out in the US, so it is left out of the picks. __Mini PCs ranked by usable memory__ | Mini PC | Price | Usable memory | Bandwidth | $ per usable GB | Biggest model at Q4 | Models that fit | | ----------------------------------------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------- | --------- | --------------- | ----------------------------------------------------------------------------------------- | --------------- | | [Mac mini M4 16GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m4-16/)Mac mini · 16 GB unified · discontinued (sale price) | $789B&H sale, 16GB/256GB MU9D3LL/A (was $899 Apple list; discontinued at Apple, replaced by M6; was $799 on 2026-10-29), [source](https://www.bhphotovideo.com/c/product/1859258-REG/apple%5Fmu9d3ll%5Fa%5Fmac%5Fmini%5Fm4%5F10c%5F10cgpu%5F16gb%5F256gb.html), 2026-11-05 | 12 GB | 120 GB/s | $66 | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [Mac mini M6 16GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-16/)Mac mini · 16 GB unified · in stock | $899Apple list price, M6 12C CPU/12C GPU, 16GB/256GB (MHQK4LL/A), [source](https://www.apple.com/shop/buy-mac/mac-mini/m6-chip-12-core-cpu-12-core-gpu-16gb-memory-256gb-storage), 2026-11-12 | 12 GB | 153 GB/s | $75 | [OLMo-2-1124-13B-Instruct](https://whatcanirunai.vercel.app/models/olmo-2-1124-13b-instruct/)\~10.6 GB at Q4 | 350 / 656 | | [Mac mini M6 24GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-24/)Mac mini · 24 GB unified · in stock | $1,299Apple list price, M6 12C CPU/12C GPU, 24GB/512GB (MHQM4LL/A), [source](https://www.apple.com/shop/buy-mac/mac-mini/m6-chip-12-core-cpu-12-core-gpu-24gb-memory-512gb-storage), 2026-11-12 | 18 GB | 170 GB/s | $72 | [gpt-oss-20b](https://whatcanirunai.vercel.app/models/gpt-oss-20b/)\~15.6 GB at Q4 | 393 / 656 | | [Mac mini M6 32GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-32/)Mac mini · 32 GB unified · in stock | $1,799Apple list price, M6 12C CPU/12C GPU, 32GB/1TB (MHQV4LL/A), [source](https://www.apple.com/shop/buy-mac/mac-mini/m6-chip-12-core-cpu-12-core-gpu-32gb-memory-1tb-storage), 2026-11-12 | 24 GB | 170 GB/s | $75 | [ERNIE-4.5-VL-28B-A3B-PT](https://whatcanirunai.vercel.app/models/ernie-4.5-vl-28b-a3b-pt/)\~21.5 GB at Q4 | 423 / 656 | | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)Mac mini · 64 GB unified · in stock | $3,199Apple list price, M5 Pro 18C CPU/20C GPU, 64GB/1TB (MHQU4LL/A), [source](https://www.apple.com/shop/buy-mac/mac-mini/m5-pro-chip-18-core-cpu-20-core-gpu-64gb-memory-1tb-storage), 2026-11-12 | 48 GB | 307 GB/s | $67 | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB at Q4 | 511 / 656 | | [NVIDIA DGX Spark 64GB](https://whatcanirunai.vercel.app/hardware/dgx-spark-64/)NVIDIA Grace Blackwell mini PC · 64 GB unified · in stock | $4,999NVIDIA OEM-partner starting price (Acer/ASUS/Dell/Gigabyte/HP/MSI; not sold on NVIDIA store), [source](https://blogs.nvidia.com/blog/local-ai-dgx-spark-64gb-sync/), 2026-11-05 | 48 GB | 273 GB/s | $104 | [AI21-Jamba2-Mini](https://whatcanirunai.vercel.app/models/ai21-jamba2-mini/)\~37.0 GB at Q4 | 511 / 656 | | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)AMD Ryzen AI Max+ 395 mini PC · 128 GB unified · pre-order | $3,449Framework list price, DIY Edition 128GB (pre-order, ships within 5 business days), [source](https://frame.work/products/desktop-diy-amd-aimax300/comparison), 2026-11-05 | 96 GB | 256 GB/s | $36 | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB at Q4 | 545 / 656 | | [NVIDIA DGX Spark 128GB](https://whatcanirunai.vercel.app/hardware/dgx-spark-128/)NVIDIA Grace Blackwell mini PC · 128 GB unified · reported sold out in the US | $6,950NVIDIA list price (not re-verified 2026-11-05: NVIDIA marketplace blocked/empty response; last confirmed $6,950 on marketplace listing; widely reported sold out in the US), [source](https://marketplace.nvidia.com/en-us/enterprise/personal-ai-supercomputers/dgx-spark/), 2026-10-22 | 96 GB | 273 GB/s | $72 | [Mistral-Small-4-119B-2603](https://whatcanirunai.vercel.app/models/mistral-small-4-119b-2603/)\~84.2 GB at Q4 | 545 / 656 | | [Mac Studio M5 Ultra 256GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-256/)Mac Studio · 256 GB unified · in stock | $9,499Apple list price, M5 Ultra 30C CPU/64C GPU, 256GB/1TB, [source](https://www.apple.com/shop/buy-mac/mac-studio/m5-ultra-chip-30-core-cpu-64-core-gpu-256gb-memory-1tb-storage), 2026-11-12 | 192 GB | 1200 GB/s | $49 | [Intern-S1](https://whatcanirunai.vercel.app/models/intern-s1/)\~168.8 GB at Q4 | 574 / 656 | | [Mac Studio M5 Ultra 512GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-512/)Mac Studio · 512 GB unified · in stock | $22,299Apple list price, M5 Ultra 36C CPU/80C GPU, 512GB/1TB, [source](https://www.apple.com/shop/buy-mac/mac-studio/m5-ultra-chip-36-core-cpu-80-core-gpu-512gb-memory-1tb-storage), 2026-11-12 | 384 GB | 1200 GB/s | $58 | [Qwen3-Coder-480B-A35B-Instruct](https://whatcanirunai.vercel.app/models/qwen3-coder-480b-a35b-instruct/)\~335.8 GB at Q4 | 610 / 656 | ## Which popular models run, and how fast Estimated decode speed at Q4 in tokens per second ([how we estimate speed](https://whatcanirunai.vercel.app/method/#speed)). MoE models show an upper bound only. "Tight" means it fits with short context only. __Estimated fit and decode speed at Q4 for popular models on each mini PC__ | Mini PC | [Qwen3-8B](https://whatcanirunai.vercel.app/models/qwen3-8b/)\~6.7 GB | [gpt-oss-20b](https://whatcanirunai.vercel.app/models/gpt-oss-20b/)\~15.6 GB | [Qwen3.8-27B](https://whatcanirunai.vercel.app/models/qwen3.8-27b/)\~20.4 GB | [Qwen3-30B-A3B](https://whatcanirunai.vercel.app/models/qwen3-30b-a3b/)\~22.3 GB | [Llama-3.3-70B-Instruct](https://whatcanirunai.vercel.app/models/llama-3.3-70b-instruct/)\~50.2 GB | [gpt-oss-120b](https://whatcanirunai.vercel.app/models/gpt-oss-120b/)\~82.5 GB | | ------------------------------------------------------------------------------------------------------ | ------------------------------------- | -------------------------------------------- | -------------------------------------------- | ------------------------------------------------ | ------------------------------------------------------------------ | ---------------------------------------------- | | [Mac mini M4 16GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m4-16/)12 GB usable · 120 GB/s | Fits11–19 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [Mac mini M6 16GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-16/)12 GB usable · 153 GB/s | Fits14–25 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [Mac mini M6 24GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-24/)18 GB usable · 170 GB/s | Fits15–27 tok/s | Fitsup to 62 tok/s | Won’t fit | Won’t fit | Won’t fit | Won’t fit | | [Mac mini M6 32GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-32/)24 GB usable · 170 GB/s | Fits15–27 tok/s | Fitsup to 62 tok/s | Fits5–8 tok/s | Tightup to 75 tok/s | Won’t fit | Won’t fit | | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)48 GB usable · 307 GB/s | Fits28–49 tok/s | Fitsup to 113 tok/s | Fits8–15 tok/s | Fitsup to 135 tok/s | Won’t fit | Won’t fit | | [NVIDIA DGX Spark 64GB](https://whatcanirunai.vercel.app/hardware/dgx-spark-64/)48 GB usable · 273 GB/s | Fits25–44 tok/s | Fitsup to 100 tok/s | Fits7–13 tok/s | Fitsup to 120 tok/s | Won’t fit | Won’t fit | | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)96 GB usable · 256 GB/s | Fits23–41 tok/s | Fitsup to 94 tok/s | Fits7–12 tok/s | Fitsup to 113 tok/s | Fits3–5 tok/s | Fitsup to 66 tok/s | | [NVIDIA DGX Spark 128GB](https://whatcanirunai.vercel.app/hardware/dgx-spark-128/)96 GB usable · 273 GB/s | Fits25–44 tok/s | Fitsup to 100 tok/s | Fits7–13 tok/s | Fitsup to 120 tok/s | Fits3–5 tok/s | Fitsup to 71 tok/s | | [Mac Studio M5 Ultra 256GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-256/)192 GB usable · 1200 GB/s | Fits109–193 tok/s | Fitsup to 440 tok/s | Fits32–57 tok/s | Fitsup to 528 tok/s | Fits13–22 tok/s | Fitsup to 310 tok/s | | [Mac Studio M5 Ultra 512GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-512/)384 GB usable · 1200 GB/s | Fits109–193 tok/s | Fitsup to 440 tok/s | Fits32–57 tok/s | Fitsup to 528 tok/s | Fits13–22 tok/s | Fitsup to 310 tok/s | ## Mini PC vs GPUs with the same memory The cheapest set of identical current GPUs (up to 4) whose VRAM adds up to at least the mini PC's usable memory, at list price. **GPU cost only**: add a CPU, board, RAM, PSU, case and SSD (price one in the [planner](https://whatcanirunai.vercel.app/planner/)). Speeds are estimates for the biggest popular model the mini PC fits; with several cards, speed is about one card's (see [multi-GPU](https://whatcanirunai.vercel.app/multi-gpu-for-local-llm/)). __Each mini PC against the cheapest GPUs with the same memory, with estimated speed__ | Mini PC | GPUs with the same memory | GPU cost | Reference model (Q4) | Mini PC est. tok/s | GPUs est. tok/s | | ---------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------- | -------- | --------------------------------------- | ------------------ | --------------- | | [Mac mini M6 16GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-16/)12 GB usable · $899 | [Intel Arc B580](https://whatcanirunai.vercel.app/hardware/arc-b580/)12 GB VRAM · 456 GB/s per card | $249 | [Qwen3-8B](https://whatcanirunai.vercel.app/models/qwen3-8b/) | 14–25 | 55–78 | | [Mac mini M6 24GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-24/)18 GB usable · $1,299 | 2× [Intel Arc B570](https://whatcanirunai.vercel.app/hardware/arc-b570/)20 GB VRAM · 380 GB/s per card | $438 | [gpt-oss-20b](https://whatcanirunai.vercel.app/models/gpt-oss-20b/) | up to 62 | up to 148 | | [Mac mini M6 32GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m6-32/)24 GB usable · $1,799 | 2× [Intel Arc B580](https://whatcanirunai.vercel.app/hardware/arc-b580/)24 GB VRAM · 456 GB/s per card | $498 | [Qwen3.8-27B](https://whatcanirunai.vercel.app/models/qwen3.8-27b/) | 5–8 | 16–23 | | [Mac mini M5 Pro 64GB](https://whatcanirunai.vercel.app/hardware/mac-mini-m5-pro-64/)48 GB usable · $3,199 | 3× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)48 GB VRAM · 288 GB/s per card | $987 | [Qwen3-30B-A3B](https://whatcanirunai.vercel.app/models/qwen3-30b-a3b/) | up to 135 | up to 135 | | [NVIDIA DGX Spark 64GB](https://whatcanirunai.vercel.app/hardware/dgx-spark-64/)48 GB usable · $4,999 | 3× [Radeon RX 7600 XT](https://whatcanirunai.vercel.app/hardware/rx-7600-xt/)48 GB VRAM · 288 GB/s per card | $987 | [Qwen3-30B-A3B](https://whatcanirunai.vercel.app/models/qwen3-30b-a3b/) | up to 120 | up to 135 | | [Framework Desktop (Ryzen AI Max+ 395) 128GB](https://whatcanirunai.vercel.app/hardware/framework-desktop-128/)96 GB usable · $3,449 | 3× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)96 GB VRAM · 608 GB/s per card | $2,847 | [gpt-oss-120b](https://whatcanirunai.vercel.app/models/gpt-oss-120b/) | up to 66 | up to 167 | | [NVIDIA DGX Spark 128GB](https://whatcanirunai.vercel.app/hardware/dgx-spark-128/)96 GB usable · $6,950 | 3× [Intel Arc Pro B70 32GB](https://whatcanirunai.vercel.app/hardware/arc-pro-b70/)96 GB VRAM · 608 GB/s per card | $2,847 | [gpt-oss-120b](https://whatcanirunai.vercel.app/models/gpt-oss-120b/) | up to 71 | up to 167 | | [Mac Studio M5 Ultra 256GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-256/)192 GB usable · $9,499 | 2× [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)192 GB VRAM · 1792 GB/s per card | $17,130 | [gpt-oss-120b](https://whatcanirunai.vercel.app/models/gpt-oss-120b/) | up to 310 | up to 493 | | [Mac Studio M5 Ultra 512GB](https://whatcanirunai.vercel.app/hardware/mac-studio-m5-ultra-512/)384 GB usable · $22,299 | 4× [NVIDIA RTX Pro 6000 Blackwell 96GB](https://whatcanirunai.vercel.app/hardware/rtx-pro-6000-blackwell/)384 GB VRAM · 1792 GB/s per card | $34,260 | [gpt-oss-120b](https://whatcanirunai.vercel.app/models/gpt-oss-120b/) | up to 310 | up to 493 | ## Need it portable? Laptops we track with this class of memory: [MacBook Pro M5 Max 128GB](https://whatcanirunai.vercel.app/hardware/macbook-pro-m5-max-128/) (96 GB usable, 614 GB/s, $6,699), [MacBook Pro M6 32GB](https://whatcanirunai.vercel.app/hardware/macbook-pro-m6-32/) (24 GB usable, 170 GB/s, $2,399). See the [best Mac for local LLMs](https://whatcanirunai.vercel.app/best-mac-for-local-llm/). Head-to-head: [Mac mini M5 Pro vs Framework Desktop](https://whatcanirunai.vercel.app/compare/mac-mini-m5-pro-64-vs-framework-desktop-128/) · [Framework Desktop vs DGX Spark](https://whatcanirunai.vercel.app/compare/framework-desktop-128-vs-dgx-spark-128/) · [DGX Spark vs Mac Studio](https://whatcanirunai.vercel.app/compare/dgx-spark-128-vs-mac-studio-m5-ultra-256/). Have a different mini PC? Enter its memory in the [VRAM calculator](https://whatcanirunai.vercel.app/vram-calculator/?type=unified) and choose unified memory. ## Quick answers What is the best mini PC for running local LLMs? It depends on the model size. Cheapest way in: the Mac mini M6 16GB (12 GB usable, $899). Best for 27–32B models: the Mac mini M6 32GB (24 GB usable, $1,799). Best for 70B models and gpt-oss-120b: the Framework Desktop (Ryzen AI Max+ 395) 128GB (96 GB usable, $3,449). Most memory: the Mac Studio M5 Ultra 512GB (384 GB usable, $22,299). Memory decides what fits; memory bandwidth decides speed. What is the cheapest mini PC for local AI? Of the compact desktops we track, the Mac mini M6 16GB at $899 (2026-11-12). It has about 12 GB usable and runs 350 of our 656 tracked models at Q4, mostly 8B-class and smaller. Can a mini PC run a 70B model? Yes, with 128 GB of unified memory. Llama 3.3 70B needs about 50.2 GB at Q4, more than a 64 GB machine's \~48 GB usable. The cheapest in-stock mini PC we track that fits it is the Framework Desktop (Ryzen AI Max+ 395) 128GB at $3,449\. Expect roughly 3–5 tok/s (estimate). Framework Desktop vs DGX Spark vs Mac Studio for local LLMs? Framework Desktop (Ryzen AI Max+ 395) 128GB: 96 GB usable, 256 GB/s, $3,449 (pre-order). NVIDIA DGX Spark 128GB: 96 GB usable, 273 GB/s, $6,950 (reported sold out in the US). Mac Studio M5 Ultra 256GB: 192 GB usable, 1200 GB/s, $9,499 (in stock). The 128 GB machines fit the same models; bandwidth sets speed, so the Mac Studio is fastest and largest but costs the most. Mini PC or GPU desktop for local LLMs? A mini PC gives you a lot of memory in one small, quiet box, so it fits bigger models for the money. A desktop GPU has far more bandwidth, so any model that fits in its VRAM runs several times faster. Matching a 96 GB mini PC with GPUs takes several cards plus a full PC around them. How fast are these speed estimates? They are rough decode-speed ranges: memory bandwidth divided by the model's Q4 weight size, times an efficiency band calibrated on public llama.cpp results. MoE models show an upper bound only. Real speed depends on the app, context length and settings. How much of a mini PC's memory can an LLM use? We count about 75% of unified memory as usable for model weights and cache, since the OS and other apps need the rest by default. That is a rule of thumb: some systems let you raise the GPU share. [Buying guides](https://whatcanirunai.vercel.app/best-gpu-for/) · [Best Mac](https://whatcanirunai.vercel.app/best-mac-for-local-llm/) · [Best GPU](https://whatcanirunai.vercel.app/best-gpu-for-local-llm/) · [NVIDIA vs AMD vs Mac](https://whatcanirunai.vercel.app/nvidia-vs-amd-vs-mac/) · [Tokens per second](https://whatcanirunai.vercel.app/tokens-per-second/) · [Multi-GPU](https://whatcanirunai.vercel.app/multi-gpu-for-local-llm/) --- Agent guide: https://whatcanirunai.vercel.app/llms.txt · JSON API: https://whatcanirunai.vercel.app/developers/