どこから始めますか？
手元のハードウェアか使いたいモデルから始め、導入前に容量・性能・根拠を比較します。
Select hardware to see model fit results.
Guides & methodology
Browse the complete catalogs and estimation method when you need them.
全ハードウェア21
機材ごとにメモリ予算、推論経路、現実的な上限が異なります。
Mac Studio M5 Ultra · 96 GB
The entry M5 Ultra configuration combines a 96 GB unified pool with 1.2 TB/s memory bandwidth. Apple announced it in August 2026; availability begins September 22.
ガイドを開くMac mini M6 · 24 GB
A compact current-generation Mac with 24 GB unified memory. Keep macOS, runtime, and KV-cache headroom separate from model weights.
ガイドを開くMacBook Pro M5 Max · 64 GB
A portable Apple-native tier for larger quantized models. Sustained inference still depends on thermals, runtime support, and context length.
ガイドを開くFramework Desktop · Ryzen AI Max+ 395 · 128 GB
A 128 GB Strix Halo desktop. AMD allows a large graphics-memory allocation, but usable capacity and speed depend on BIOS, driver, and runtime support.
ガイドを開くRadeon AI PRO R9700 · 32 GB
A 32 GB AMD workstation card for larger single-GPU quantized models. Check the exact operating-system and ROCm or Vulkan runtime path first.
ガイドを開くNVIDIA DGX Spark · 128 GB
A compact local AI system with 128 GB coherent memory. It can hold large low-bit models, but decode speed and software support remain model-specific.
ガイドを開くMac Studio M3 Ultra · 96 GB
A quiet single-box option for 70B-class Q4 models and Apple-native MLX workflows, with memory shared by macOS and applications.
ガイドを開くMac Studio M3 Ultra · 256 GB
Enough capacity for many 100B-plus low-bit models, provided the architecture is supported and optimized by the selected runtime.
ガイドを開くGeForce RTX 5090 · 32 GB
A fast single-GPU tier for 27B-class low-bit models, with limited remaining space for very long context or multiple concurrent users.
ガイドを開くGeForce RTX 4090 · 24 GB
A 27B Q4 checkpoint can be a tight fit. Leave room for runtime overhead and KV cache instead of filling every gigabyte with weights.
ガイドを開く8 GB CPU laptop
Small text models with short context. Expect patient, private inference rather than speed.
ガイドを開く16 GB CPU desktop
Comfortable with 3B–8B Q4 models; 12B is possible only with reduced context and patience.
ガイドを開くApple silicon · 16 GB
A polished 4B–8B local experience when the OS and apps have enough headroom.
ガイドを開くNVIDIA · 8 GB
The mainstream 4B–8B tier. Some 12B INT4 builds fit tightly with modest context.
ガイドを開くNVIDIA · 12 GB
Strong 8B–14B Q4 territory for a single user.
ガイドを開くNVIDIA · 16 GB
14B models are comfortable; 20B-class low-bit MoE models are a tight upper edge.
ガイドを開くApple silicon · 32 GB
24B–32B Q4 models become practical for one local user.
ガイドを開くNVIDIA · 24 GB
The consumer sweet spot for 27B and 30B-A3B INT4/Q4 models.
ガイドを開くApple silicon · 64 GB
32B models with long context or 70B Q4 with conservative headroom.
ガイドを開くWorkstation · 48 GB
70B Q4 serving begins here, with limited room for cache and concurrency.
ガイドを開くDatacenter · 80 GB
Large MoE models and high-throughput servers, still bounded by quantization and cache.
ガイドを開く全モデル24
Qwen3.8 Flash Next
English- 推定メモリ
- ~108 GB
- コンテキスト
- 262,144 native; 1M extended
Ling-3.0-flash
English- 推定メモリ
- ~74.4 GB
- コンテキスト
- Check selected runtime and checkpoint
Nex-N2.5-mini
English- 推定メモリ
- ~21 GB
- コンテキスト
- 262,144 (config max_position_embeddings)
Qwen3.6 27B
- 推定メモリ
- ~18.5 GB
- コンテキスト
- 262K native
Qwen3.8 27B
- 推定メモリ
- ~19.5 GB
- コンテキスト
- 262K native
Qwen3.8 Flash Next REAP-288
- 推定メモリ
- ~40.57 GB
- コンテキスト
- 7.6K recommended on measured 48 GB Mac
Qwen3.6 35B-A3B
- 推定メモリ
- ~25 GB
- コンテキスト
- 262K native
DeepSeek V4 Flash
- 推定メモリ
- ~176 GB
- コンテキスト
- 1M native
Gemma 3 1B
- 推定メモリ
- ~1.4 GB
- コンテキスト
- 32K
Llama 3.2 3B
- 推定メモリ
- ~2.8 GB
- コンテキスト
- 128K
Qwen3 4B
- 推定メモリ
- ~3.6 GB
- コンテキスト
- 32K+
Gemma 3 4B
- 推定メモリ
- ~4.2 GB
- コンテキスト
- 128K
Qwen3 8B
- 推定メモリ
- ~6.8 GB
- コンテキスト
- 32K+
Gemma 3 12B
- 推定メモリ
- ~9.4 GB
- コンテキスト
- 128K
Qwen3 14B
- 推定メモリ
- ~11.2 GB
- コンテキスト
- 32K+
gpt-oss-20b
- 推定メモリ
- ~16 GB
- コンテキスト
- 128K
Devstral Small 2 24B
- 推定メモリ
- ~18.5 GB
- コンテキスト
- 256K
Gemma 3 27B
- 推定メモリ
- ~20.5 GB
- コンテキスト
- 128K
Qwen3 30B-A3B
- 推定メモリ
- ~22.5 GB
- コンテキスト
- 32K+
Qwen3 32B
- 推定メモリ
- ~24.5 GB
- コンテキスト
- 32K+
Llama 3.1 70B
- 推定メモリ
- ~46 GB
- コンテキスト
- 128K
Llama 4 Scout
- 推定メモリ
- ~78 GB
- コンテキスト
- 10M advertised
gpt-oss-120b
- 推定メモリ
- ~80 GB
- コンテキスト
- 128K
Qwen-Image 2.1
English- 推定メモリ
- ~33.1 GB
- コンテキスト
- Not a token context: one prompt plus up to 10 reference images
目的に合う推論レイヤー15
Ollama
One-command local chat and app integration
ガイドを開くllama.cpp
GGUF models, CPU/GPU offload, embedded and unusual hardware
ガイドを開くLM Studio
Discovering, downloading and testing models without a terminal
ガイドを開くMLX LM
Native Apple silicon inference, experimentation and fine-tuning
ガイドを開くoMLX
Apple-silicon serving with continuous batching, memory guards, and supported SSD-backed paths
ガイドを開くvLLM
Linux GPU servers, concurrency and OpenAI-compatible production APIs
ガイドを開くTensorRT-LLM
Maximum NVIDIA throughput after engine tuning
ガイドを開くSGLang
Multi-request GPU serving with prefix caching and speculative decoding
ガイドを開くLocalAI
One OpenAI-compatible endpoint for text, vision, speech and image models, GPU optional
ガイドを開くLemonade
Ryzen AI, Radeon and Strix Halo PCs, including the XDNA2 NPU, behind one local API
ガイドを開くOpenVINO Model Server
Serving OpenVINO IR models on Intel CPUs, Arc GPUs and NPUs with one API
ガイドを開くKoboldCpp
One-file GGUF chat on Windows, macOS or Linux with no Python install
ガイドを開くllamafile
Distributing a whole GGUF model as one runnable executable on Linux, macOS, Windows and BSD
ガイドを開くKTransformers
Very large MoE models with hot experts on the GPU and cold experts in system RAM
ガイドを開くUnsloth
Running GGUF or MLX models from a desktop app that also trains and exports them
ガイドを開く推定方法
公式サイズと量子化を基にOSと実行環境の余裕を確保し、KVキャッシュは追加負荷として扱います。MoEは総パラメータで判定します。
総パラメータ、MoEの有効パラメータ、量子化サイズ、対応環境、実測性能を分けて扱います。
- Memory fit is an estimate, not a speed measurement.
- Size MoE by total parameters; shared and distributed memory need additional checks.
- Vendor claims, community measurements, runtime conditions and unknowns are disclosed separately.
Catalog data updated: 2026-10-03
40GBをダウンロードする前に
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.
How do you report speed?
Exact device-and-model measurements are labeled measured. Relative manufacturer figures are labeled vendor claims. Every other combination stays unknown instead of receiving an invented tokens-per-second estimate.