01
Qwen3 30B-A3BKetat
ollama run qwen3:30b-a3bPanduan lapangan AI privat
Jalankan model yang tepat.
Di perangkat yang sudah Anda punya.
Pilih platform dan memori. Dapatkan kecocokan konservatif, server yang tepat, dan perintah yang bisa dijalankan.
Primary-source linked16 languagesBukti diperbarui 2026-08-31
8 GB
24 GB
80GB
ollama runqwen3:8b
LOCAL / 24G
MEMORY MAP01—80 GB
RECOMMENDED ENVELOPE6 models
Lega Ketat
02
Gemma 3 27BKetat
ollama run gemma3:27b03
Devstral Small 2 24BKetat
ollama run <model-tag>04
gpt-oss-20bLega
ollama run gpt-oss:20b05
Qwen3 14BLega
ollama run qwen3:14b06
Gemma 3 12BLega
ollama run gemma3:12b02 / EVIDENCE FIRST
Peta, bukan peringkat
Kami memisahkan parameter total, MoE aktif, ukuran kuantisasi, dukungan runtime, dan kinerja terukur.
Weight fit ≠ speedTotal ≠ active parametersDocs ≠ benchmark
03 / HARDWARE MAP
Dari laptop hingga akselerator 80 GB
Setiap kelas punya anggaran, jalur runtime, dan batas realistis sendiri.
8GB
Everyday laptop
8 GB CPU laptop
Small text models with short context. Expect patient, private inference rather than speed.
Buka panduan16GB
No discrete GPU
16 GB CPU desktop
Comfortable with 3B–8B Q4 models; 12B is possible only with reduced context and patience.
Buka panduan16GB
M-series notebook
Apple silicon · 16 GB
A polished 4B–8B local experience when the OS and apps have enough headroom.
Buka panduan8GB
RTX 4060 class
NVIDIA · 8 GB
The mainstream 4B–8B tier. Some 12B INT4 builds fit tightly with modest context.
Buka panduan12GB
RTX 3060 / 4070 class
NVIDIA · 12 GB
Strong 8B–14B Q4 territory for a single user.
Buka panduan16GB
RTX 4080 / 5080 class
NVIDIA · 16 GB
14B models are comfortable; 20B-class low-bit MoE models are a tight upper edge.
Buka panduan04 / MODEL FIELD NOTES
Model open-weight perwakilan
Google
Gemma 3 1B
1BINT4 / GGUF Q4
- Estimasi memori
- ~1.4 GB
- Konteks
- 32K
Alibaba Qwen
Qwen3 4B
4BGGUF Q4
- Estimasi memori
- ~3.6 GB
- Konteks
- 32K+
Alibaba Qwen
Qwen3 8B
8BGGUF Q4
- Estimasi memori
- ~6.8 GB
- Konteks
- 32K+
OpenAI
gpt-oss-20b
21B3.6B active · MXFP4
- Estimasi memori
- ~16 GB
- Konteks
- 128K
Alibaba Qwen
Qwen3 30B-A3B
30B3B active · GGUF Q4 MoE
- Estimasi memori
- ~22.5 GB
- Konteks
- 32K+
05 / SERVING LAYER
Pilih lapisan inferensi
Fastest first run
Ollama
One-command local chat and app integration
Buka panduanWidest hardware reach
llama.cpp
GGUF models, CPU/GPU offload, embedded and unusual hardware
Buka panduanBest visual desktop
LM Studio
Discovering, downloading and testing models without a terminal
Buka panduanApple silicon specialist
MLX LM
Native Apple silicon inference, experimentation and fine-tuning
Buka panduanThroughput server
vLLM
Linux GPU servers, concurrency and OpenAI-compatible production APIs
Buka panduanNVIDIA optimized
TensorRT-LLM
Maximum NVIDIA throughput after engine tuning
Buka panduan06 / METHOD
Cara estimasi dibuat
Kami memakai ukuran dan kuantisasi terdokumentasi, menyisakan ruang untuk sistem, dan menghitung cache sebagai tambahan. MoE memakai parameter total.
CONSERVATIVE FITweights + runtime + cache
total params×bits / 8+headroom
No universal tokens/s claims.07 / FAQ
Sebelum mengunduh 40 GB
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.