01
Qwen3 30B-A3B勉強適配
ollama run qwen3:30b-a3b你的私有 AI 部署地圖
選對模型，
跑在你已有的硬體上。
選擇平台與記憶體，我們提供保守的模型適配、合適的推論服務，以及真正可執行的啟動指令。
Primary-source linked16 languages證據更新於 2026-08-31
8 GB
24 GB
80GB
ollama runqwen3:8b
LOCAL / 24G
MEMORY MAP01—80 GB
RECOMMENDED ENVELOPE6 models
餘量充足 勉強適配
02
Gemma 3 27B勉強適配
ollama run gemma3:27b03
ollama run <model-tag>04
gpt-oss-20b餘量充足
ollama run gpt-oss:20b05
Qwen3 14B餘量充足
ollama run qwen3:14b06
Gemma 3 12B餘量充足
ollama run gemma3:12b02 / EVIDENCE FIRST
這是地圖，不是排行榜
我們區分總參數、MoE 啟用參數、量化權重、執行環境支援與實測效能；未知速度保持未知。
Weight fit ≠ speedTotal ≠ active parametersDocs ≠ benchmark
03 / HARDWARE MAP
從安靜的筆電到 80GB 加速卡
每種設備都有不同的記憶體預算、推論路徑與誠實上限。
8GB
Everyday laptop
8 GB CPU laptop
Small text models with short context. Expect patient, private inference rather than speed.
開啟指南16GB
No discrete GPU
16 GB CPU desktop
Comfortable with 3B–8B Q4 models; 12B is possible only with reduced context and patience.
開啟指南16GB
M-series notebook
Apple silicon · 16 GB
A polished 4B–8B local experience when the OS and apps have enough headroom.
開啟指南8GB
RTX 4060 class
NVIDIA · 8 GB
The mainstream 4B–8B tier. Some 12B INT4 builds fit tightly with modest context.
開啟指南12GB
RTX 3060 / 4070 class
NVIDIA · 12 GB
Strong 8B–14B Q4 territory for a single user.
開啟指南16GB
RTX 4080 / 5080 class
NVIDIA · 16 GB
14B models are comfortable; 20B-class low-bit MoE models are a tight upper edge.
開啟指南04 / MODEL FIELD NOTES
代表性開放權重模型
05 / SERVING LAYER
為工作選擇推論層
Fastest first run
Ollama
One-command local chat and app integration
開啟指南Widest hardware reach
llama.cpp
GGUF models, CPU/GPU offload, embedded and unusual hardware
開啟指南Best visual desktop
LM Studio
Discovering, downloading and testing models without a terminal
開啟指南Apple silicon specialist
MLX LM
Native Apple silicon inference, experimentation and fine-tuning
開啟指南Throughput server
vLLM
Linux GPU servers, concurrency and OpenAI-compatible production APIs
開啟指南NVIDIA optimized
TensorRT-LLM
Maximum NVIDIA throughput after engine tuning
開啟指南06 / METHOD
估算方式
依官方模型大小與量化格式估算，為系統和執行環境預留空間，並將上下文快取視為額外成本。MoE 使用總參數。
CONSERVATIVE FITweights + runtime + cache
total params×bits / 8+headroom
No universal tokens/s claims.07 / FAQ
下載 40GB 前先問清楚
Does a 24 GB GPU run a 30B model?+
Often at Q4/INT4 with a conservative context. Qwen3 30B-A3B and Gemma 3 27B are representative fits, but cache and runtime overhead still matter.
Are active MoE parameters the memory requirement?+
No. Active parameters affect compute per token; total parameters still need to be stored in memory or offloaded.
Which runtime should a beginner choose?+
Ollama for a terminal-first setup or LM Studio for a visual desktop. llama.cpp is the portable fallback; vLLM is for higher-throughput GPU serving.
Do you benchmark speed?+
Not yet. Launch recommendations are transparent memory-fit estimates backed by primary documentation. We do not invent tokens-per-second numbers.