# OnPrem World (地端世界)

> Latest prices, news and a forum for running AI on your own hardware: which models run on which GPU or Mac, and how fast. 地端 AI 的最新價格、新聞與論壇：哪個模型能在哪張卡、哪台 Mac 上跑，跑多快，一站查清楚。 Data updated 2026-10-14. Licence CC BY 4.0.

Every page has a Markdown version: append `.md` to its URL. Pages exist in English (/en) and Traditional Chinese (/zh-tw).

## Start here
- [Which LLMs can your hardware run?](https://onpremworld.com/en.md)
- [Can I run it? (calculator + API)](https://onpremworld.com/en/can-i-run.md)
- [How the estimates work](https://onpremworld.com/en/methodology.md)
- [你的硬體能跑哪些 LLM？](https://onpremworld.com/zh-tw.md)

## Hardware
- [All hardware](https://onpremworld.com/en/hardware.md)
- [GeForce RTX 3060 (12 GB)](https://onpremworld.com/en/hardware/rtx-3060-12gb.md)
- [GeForce RTX 3090](https://onpremworld.com/en/hardware/rtx-3090.md)
- [GeForce RTX 4060 Ti 16GB](https://onpremworld.com/en/hardware/rtx-4060-ti-16gb.md)
- [GeForce RTX 4080 SUPER](https://onpremworld.com/en/hardware/rtx-4080-super.md)
- [GeForce RTX 4090](https://onpremworld.com/en/hardware/rtx-4090.md)
- [GeForce RTX 5060 Ti 16GB](https://onpremworld.com/en/hardware/rtx-5060-ti-16gb.md)
- [GeForce RTX 5070](https://onpremworld.com/en/hardware/rtx-5070.md)
- [GeForce RTX 5070 Ti](https://onpremworld.com/en/hardware/rtx-5070-ti.md)
- [GeForce RTX 5080](https://onpremworld.com/en/hardware/rtx-5080.md)
- [GeForce RTX 5090](https://onpremworld.com/en/hardware/rtx-5090.md)
- [Radeon RX 7900 XT](https://onpremworld.com/en/hardware/amd-rx-7900-xt.md)
- [Radeon RX 7900 XTX](https://onpremworld.com/en/hardware/amd-rx-7900-xtx.md)
- [Radeon RX 9060 XT 16GB](https://onpremworld.com/en/hardware/amd-rx-9060-xt-16gb.md)
- [Radeon RX 9070 XT](https://onpremworld.com/en/hardware/amd-rx-9070-xt.md)
- [Arc B580](https://onpremworld.com/en/hardware/intel-arc-b580.md)
- [RTX A6000](https://onpremworld.com/en/hardware/nvidia-rtx-a6000.md)
- [RTX 6000 Ada Generation](https://onpremworld.com/en/hardware/nvidia-rtx-6000-ada.md)
- [RTX PRO 5000 72GB Blackwell](https://onpremworld.com/en/hardware/nvidia-rtx-pro-5000-blackwell-72gb.md)
- [RTX PRO 6000 Blackwell Workstation Edition](https://onpremworld.com/en/hardware/nvidia-rtx-pro-6000-blackwell.md)
- [Radeon AI PRO R9700](https://onpremworld.com/en/hardware/amd-radeon-ai-pro-r9700.md)
- [Arc Pro B60](https://onpremworld.com/en/hardware/intel-arc-pro-b60.md)
- [Arc Pro B70](https://onpremworld.com/en/hardware/intel-arc-pro-b70.md)
- [M4 Pro](https://onpremworld.com/en/hardware/apple-m4-pro.md)
- [M4 Max (14-core CPU / 32-core GPU)](https://onpremworld.com/en/hardware/apple-m4-max-32c-gpu.md)
- [M4 Max (16-core CPU / 40-core GPU)](https://onpremworld.com/en/hardware/apple-m4-max-40c-gpu.md)
- [M3 Ultra](https://onpremworld.com/en/hardware/apple-m3-ultra.md)
- [M5](https://onpremworld.com/en/hardware/apple-m5.md)
- [M5 Pro](https://onpremworld.com/en/hardware/apple-m5-pro.md)
- [M5 Max (18-core CPU / 32-core GPU)](https://onpremworld.com/en/hardware/apple-m5-max-32c-gpu.md)
- [M5 Max (18-core CPU / 40-core GPU)](https://onpremworld.com/en/hardware/apple-m5-max-40c-gpu.md)
- [M5 Ultra](https://onpremworld.com/en/hardware/apple-m5-ultra.md)
- [M6 (16GB configuration, 153 GB/s)](https://onpremworld.com/en/hardware/apple-m6-16gb.md)
- [M6 (24GB/32GB configurations, 170 GB/s)](https://onpremworld.com/en/hardware/apple-m6-24-32gb.md)
- [DGX Spark](https://onpremworld.com/en/hardware/nvidia-dgx-spark.md)
- [RTX Spark (N1X) laptops & compact desktops](https://onpremworld.com/en/hardware/nvidia-rtx-spark-n1x.md)
- [Ryzen AI Max+ 395 (Strix Halo) systems](https://onpremworld.com/en/hardware/amd-ryzen-ai-max-plus-395.md)
- [Ryzen AI Max+ PRO 495 (Gorgon Halo) systems](https://onpremworld.com/en/hardware/amd-ryzen-ai-max-plus-pro-495.md)

## Models
- [All models](https://onpremworld.com/en/models.md)
- [Llama 3.1 8B Instruct](https://onpremworld.com/en/models/llama-3.1-8b.md)
- [Llama 3.2 3B Instruct](https://onpremworld.com/en/models/llama-3.2-3b.md)
- [Llama 3.3 70B Instruct](https://onpremworld.com/en/models/llama-3.3-70b.md)
- [Llama 4 Scout 17B-16E Instruct](https://onpremworld.com/en/models/llama-4-scout.md)
- [Qwen3-8B](https://onpremworld.com/en/models/qwen3-8b.md)
- [Qwen3-32B](https://onpremworld.com/en/models/qwen3-32b.md)
- [Qwen3-30B-A3B (2507)](https://onpremworld.com/en/models/qwen3-30b-a3b.md)
- [Qwen3-Coder-30B-A3B-Instruct](https://onpremworld.com/en/models/qwen3-coder-30b-a3b.md)
- [Qwen3-Coder-Next (80B-A3B)](https://onpremworld.com/en/models/qwen3-coder-next.md)
- [Qwen3.5-9B](https://onpremworld.com/en/models/qwen3.5-9b.md)
- [Qwen3.5-27B](https://onpremworld.com/en/models/qwen3.5-27b.md)
- [Qwen3.5-35B-A3B](https://onpremworld.com/en/models/qwen3.5-35b-a3b.md)
- [Qwen3.5-122B-A10B](https://onpremworld.com/en/models/qwen3.5-122b-a10b.md)
- [Qwen3.6-27B](https://onpremworld.com/en/models/qwen3.6-27b.md)
- [Qwen3.6-35B-A3B](https://onpremworld.com/en/models/qwen3.6-35b-a3b.md)
- [Qwen3.8-27B](https://onpremworld.com/en/models/qwen3.8-27b.md)
- [Qwen3.8-Flash-Next](https://onpremworld.com/en/models/qwen3.8-flash-next.md)
- [Gemma 3 12B IT](https://onpremworld.com/en/models/gemma-3-12b.md)
- [Gemma 3 27B IT](https://onpremworld.com/en/models/gemma-3-27b.md)
- [Gemma 4 E4B IT](https://onpremworld.com/en/models/gemma-4-e4b.md)
- [Gemma 4 12B Unified IT](https://onpremworld.com/en/models/gemma-4-12b.md)
- [Gemma 4 26B A4B IT](https://onpremworld.com/en/models/gemma-4-26b-a4b.md)
- [Gemma 4 31B IT](https://onpremworld.com/en/models/gemma-4-31b.md)
- [Mistral Small 3.2 24B Instruct (2506)](https://onpremworld.com/en/models/mistral-small-3.2-24b.md)
- [Magistral Small 1.2 (2509)](https://onpremworld.com/en/models/magistral-small-2509.md)
- [Devstral Small 2 24B Instruct (2512)](https://onpremworld.com/en/models/devstral-small-2-24b.md)
- [Mistral Small 4 119B (2603)](https://onpremworld.com/en/models/mistral-small-4-119b.md)
- [DeepSeek-R1-0528-Qwen3-8B](https://onpremworld.com/en/models/deepseek-r1-0528-qwen3-8b.md)
- [DeepSeek-R1-Distill-Qwen-32B](https://onpremworld.com/en/models/deepseek-r1-distill-qwen-32b.md)
- [DeepSeek-R1-0528 (671B)](https://onpremworld.com/en/models/deepseek-r1-0528.md)
- [DeepSeek-V4-Flash](https://onpremworld.com/en/models/deepseek-v4-flash.md)
- [gpt-oss-20b](https://onpremworld.com/en/models/gpt-oss-20b.md)
- [gpt-oss-120b](https://onpremworld.com/en/models/gpt-oss-120b.md)
- [Phi-4 (14B)](https://onpremworld.com/en/models/phi-4.md)
- [GLM-4.5-Air](https://onpremworld.com/en/models/glm-4.5-air.md)
- [GLM-4.7-Flash (30B-A3B)](https://onpremworld.com/en/models/glm-4.7-flash.md)
- [Kimi K2.6](https://onpremworld.com/en/models/kimi-k2.6.md)

## Prices
- [GPU and Mac prices, Taiwan (NT$) and US (US$)](https://onpremworld.com/en/prices.md): latest price per shop with URL and date, price per GB, value score
- [顯示卡與 Mac 價格](https://onpremworld.com/zh-tw/prices.md)
- [Download all prices (JSON, CC BY 4.0)](https://onpremworld.com/data/prices/observations.json)
- [Prices API example](https://onpremworld.com/api/v1/prices?hardware=rtx-5090)

## News
- [Local AI news](https://onpremworld.com/en/news.md): model, runtime and hardware releases with a short summary and the official source, latest 2026-10-13
- [地端 AI 新聞](https://onpremworld.com/zh-tw/news.md)
- RSS: https://onpremworld.com/feed.xml (English), https://onpremworld.com/zh-tw/feed.xml (中文)
- [NVIDIA RTX Spark laptops with up to 128 GB of unified memory go on sale October 16](https://onpremworld.com/en/news/2026-10-nvidia-rtx-spark-laptops.md) (2026-10-13)
- [AMD's next Ryzen AI Halo: Ryzen AI Max+ PRO 495 with up to 192 GB of unified memory](https://onpremworld.com/en/news/2026-10-amd-ryzen-ai-halo-max-plus-pro-495.md) (2026-10-12)
- [Open WebUI 0.12 adds two-factor sign-in, shared chats you can reply in, and realtime voice](https://onpremworld.com/en/news/2026-10-open-webui-0-12-two-factor-realtime-voice.md) (2026-10-10)
- [Ollama 0.40.2 upgrades older model downloads in the background and keeps backups](https://onpremworld.com/en/news/2026-10-ollama-0-40-2-background-model-upgrades.md) (2026-10-08)
- [EmbeddingGemma 2: a 740M multimodal embedding model for local search and RAG](https://onpremworld.com/en/news/2026-10-embeddinggemma-2.md) (2026-10-06)
- [llama.cpp v0.6.0 adds GLM-5.3-Flash, a decision-model API and faster Metal kernels](https://onpremworld.com/en/news/2026-10-llama-cpp-v0-6-0.md) (2026-10-05)
- [vLLM v0.31 adds `vllm preload` for fast restarts and gates per-request multimodal options](https://onpremworld.com/en/news/2026-10-vllm-v0-31-preload.md) (2026-10-05)
- [Ollama 0.35 adds decision models that answer with probabilities instead of text](https://onpremworld.com/en/news/2026-09-ollama-0-35-decision-models.md) (2026-09-28)
- [Ollama 0.40 runs supported models on MLX by default on Apple Silicon](https://onpremworld.com/en/news/2026-09-ollama-0-40-mlx-default-apple-silicon.md) (2026-09-25)
- [Mac mini with M6 and Mac Studio with M5 Ultra go on sale, from US$899 and US$5,499](https://onpremworld.com/en/news/2026-09-mac-mini-m6-mac-studio-m5-ultra-on-sale.md) (2026-09-22)

## Guides
- [All guides](https://onpremworld.com/en/guides.md) · [教學總覽](https://onpremworld.com/zh-tw/guides.md)
- [How to run an LLM on your own computer with Ollama (Windows, macOS, Linux)](https://onpremworld.com/en/guides/install-ollama.md): Install Ollama, pick a model that fits your memory, and run it with one command. Verified install steps for Windows, macOS and Linux, plus context length, GPU checks, the local API and local-only mode.
- [Choosing a GPU for local LLMs: VRAM vs memory bandwidth](https://onpremworld.com/en/guides/choosing-a-gpu.md): VRAM decides which models fit; memory bandwidth decides how fast they answer. What 12, 16, 24 and 32 GB cards can run, with estimates from our hardware and model data.
- [Running LLMs on a Mac: unified memory, how much RAM to buy, MLX vs llama.cpp](https://onpremworld.com/en/guides/mac-local-llm.md): Any Apple Silicon Mac can run local LLMs because the GPU shares the Mac's memory. How much of it the GPU can use, which models fit in 16 to 512 GB, why bandwidth decides speed, and when to use MLX or llama.cpp.
- [Quantization explained: Q4_K_M vs Q8_0 vs FP16, and the memory and quality trade-offs](https://onpremworld.com/en/guides/quantization-explained.md): What the quantization names mean, how much memory each level saves, and how much quality it costs, with published measurements from llama.cpp and Qwen and memory figures from our model data.
- [Ollama vs LM Studio vs llama.cpp vs vLLM: which runtime for which job](https://onpremworld.com/en/guides/ollama-vs-lm-studio-vs-llama-cpp-vs-vllm.md): Four ways to run open models locally, compared on platforms, model formats, APIs, licences and multi-user serving, with verified install and start commands for each as of October 2026.
- [A private ChatGPT for your office: Open WebUI + Ollama on your own hardware](https://onpremworld.com/en/guides/private-chatgpt-open-webui.md): Run a ChatGPT-style assistant for your team on a server you own: hardware sizing from our data, verified Docker setup for Open WebUI and Ollama, user accounts, document Q&A, a checklist for keeping data in-house, and the licences involved.
- [Ollama 安裝教學：在自己的電腦跑 LLM（Windows／macOS／Linux）](https://onpremworld.com/zh-tw/guides/install-ollama.md)
- [本地跑 LLM 要買什麼顯示卡？VRAM 與記憶體頻寬怎麼選](https://onpremworld.com/zh-tw/guides/choosing-a-gpu.md)
- [用 Mac 跑本地 LLM：統一記憶體怎麼算、要買多少 RAM、MLX 還是 llama.cpp](https://onpremworld.com/zh-tw/guides/mac-local-llm.md)
- [量化是什麼？Q4_K_M、Q8_0、FP16 差在哪，記憶體與品質怎麼取捨](https://onpremworld.com/zh-tw/guides/quantization-explained.md)
- [Ollama、LM Studio、llama.cpp、vLLM 怎麼選？本地 LLM 執行工具比較](https://onpremworld.com/zh-tw/guides/ollama-vs-lm-studio-vs-llama-cpp-vs-vllm.md)
- [打造公司內部的私有 ChatGPT：Open WebUI＋Ollama 地端部署教學](https://onpremworld.com/zh-tw/guides/private-chatgpt-open-webui.md)

## API and agents
- [OpenAPI description](https://onpremworld.com/openapi.json): read-only JSON API, no key
- [Fit check example](https://onpremworld.com/api/v1/fit?model=qwen3-32b&hardware=rtx-4090&context=8192)
- [Latest news as JSON](https://onpremworld.com/api/v1/news): `?category=model-release|runtime-release|hardware|price|other`
- MCP server (Streamable HTTP): https://onpremworld.com/mcp