어디서 시작하시나요?
보유한 하드웨어나 원하는 모델에서 시작해 배포 전에 용량, 성능, 근거를 비교하세요.
Select hardware to see model fit results.
Guides & methodology
Browse the complete catalogs and estimation method when you need them.
모든 하드웨어21
장치마다 메모리 예산, 추론 경로, 현실적인 상한이 다릅니다.
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.
가이드 열기Mac 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.
가이드 열기MacBook 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.
가이드 열기Framework 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.
가이드 열기Radeon 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.
가이드 열기NVIDIA 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.
가이드 열기Mac 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.
가이드 열기Mac 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.
가이드 열기GeForce 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.
가이드 열기GeForce 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.
가이드 열기8 GB CPU laptop
Small text models with short context. Expect patient, private inference rather than speed.
가이드 열기16 GB CPU desktop
Comfortable with 3B–8B Q4 models; 12B is possible only with reduced context and patience.
가이드 열기Apple silicon · 16 GB
A polished 4B–8B local experience when the OS and apps have enough headroom.
가이드 열기NVIDIA · 8 GB
The mainstream 4B–8B tier. Some 12B INT4 builds fit tightly with modest context.
가이드 열기NVIDIA · 12 GB
Strong 8B–14B Q4 territory for a single user.
가이드 열기NVIDIA · 16 GB
14B models are comfortable; 20B-class low-bit MoE models are a tight upper edge.
가이드 열기Apple silicon · 32 GB
24B–32B Q4 models become practical for one local user.
가이드 열기NVIDIA · 24 GB
The consumer sweet spot for 27B and 30B-A3B INT4/Q4 models.
가이드 열기Apple silicon · 64 GB
32B models with long context or 70B Q4 with conservative headroom.
가이드 열기Workstation · 48 GB
70B Q4 serving begins here, with limited room for cache and concurrency.
가이드 열기Datacenter · 80 GB
Large MoE models and high-throughput servers, still bounded by quantization and cache.
가이드 열기모든 모델24
Qwen3.8 Flash Next
English- 추정 메모리
- ~108 GB
- 컨텍스트
- 262,144 native; 1M extended
Ling-3.0-flash
English- 추정 메모리
- ~74.4 GB
- 컨텍스트
- Check selected runtime and checkpoint
Nex-N2.5-mini
English- 추정 메모리
- ~21 GB
- 컨텍스트
- 262,144 (config max_position_embeddings)
Qwen3.6 27B
- 추정 메모리
- ~18.5 GB
- 컨텍스트
- 262K native
Qwen3.8 27B
- 추정 메모리
- ~19.5 GB
- 컨텍스트
- 262K native
Qwen3.8 Flash Next REAP-288
- 추정 메모리
- ~40.57 GB
- 컨텍스트
- 7.6K recommended on measured 48 GB Mac
Qwen3.6 35B-A3B
- 추정 메모리
- ~25 GB
- 컨텍스트
- 262K native
DeepSeek V4 Flash
- 추정 메모리
- ~176 GB
- 컨텍스트
- 1M native
Gemma 3 1B
- 추정 메모리
- ~1.4 GB
- 컨텍스트
- 32K
Llama 3.2 3B
- 추정 메모리
- ~2.8 GB
- 컨텍스트
- 128K
Qwen3 4B
- 추정 메모리
- ~3.6 GB
- 컨텍스트
- 32K+
Gemma 3 4B
- 추정 메모리
- ~4.2 GB
- 컨텍스트
- 128K
Qwen3 8B
- 추정 메모리
- ~6.8 GB
- 컨텍스트
- 32K+
Gemma 3 12B
- 추정 메모리
- ~9.4 GB
- 컨텍스트
- 128K
Qwen3 14B
- 추정 메모리
- ~11.2 GB
- 컨텍스트
- 32K+
gpt-oss-20b
- 추정 메모리
- ~16 GB
- 컨텍스트
- 128K
Devstral Small 2 24B
- 추정 메모리
- ~18.5 GB
- 컨텍스트
- 256K
Gemma 3 27B
- 추정 메모리
- ~20.5 GB
- 컨텍스트
- 128K
Qwen3 30B-A3B
- 추정 메모리
- ~22.5 GB
- 컨텍스트
- 32K+
Qwen3 32B
- 추정 메모리
- ~24.5 GB
- 컨텍스트
- 32K+
Llama 3.1 70B
- 추정 메모리
- ~46 GB
- 컨텍스트
- 128K
Llama 4 Scout
- 추정 메모리
- ~78 GB
- 컨텍스트
- 10M advertised
gpt-oss-120b
- 추정 메모리
- ~80 GB
- 컨텍스트
- 128K
Qwen-Image 2.1
English- 추정 메모리
- ~33.1 GB
- 컨텍스트
- Not a token context: one prompt plus up to 10 reference images
목적에 맞는 서빙 레이어15
Ollama
One-command local chat and app integration
가이드 열기llama.cpp
GGUF models, CPU/GPU offload, embedded and unusual hardware
가이드 열기LM Studio
Discovering, downloading and testing models without a terminal
가이드 열기MLX LM
Native Apple silicon inference, experimentation and fine-tuning
가이드 열기oMLX
Apple-silicon serving with continuous batching, memory guards, and supported SSD-backed paths
가이드 열기vLLM
Linux GPU servers, concurrency and OpenAI-compatible production APIs
가이드 열기TensorRT-LLM
Maximum NVIDIA throughput after engine tuning
가이드 열기SGLang
Multi-request GPU serving with prefix caching and speculative decoding
가이드 열기LocalAI
One OpenAI-compatible endpoint for text, vision, speech and image models, GPU optional
가이드 열기Lemonade
Ryzen AI, Radeon and Strix Halo PCs, including the XDNA2 NPU, behind one local API
가이드 열기OpenVINO Model Server
Serving OpenVINO IR models on Intel CPUs, Arc GPUs and NPUs with one API
가이드 열기KoboldCpp
One-file GGUF chat on Windows, macOS or Linux with no Python install
가이드 열기llamafile
Distributing a whole GGUF model as one runnable executable on Linux, macOS, Windows and BSD
가이드 열기KTransformers
Very large MoE models with hot experts on the GPU and cold experts in system RAM
가이드 열기Unsloth
Running GGUF or MLX models from a desktop app that also trains and exports them
가이드 열기추정 방식
공식 모델 크기와 양자화를 기준으로 OS와 런타임 여유를 두고, 컨텍스트 캐시는 별도 비용으로 봅니다. MoE는 총 파라미터를 사용합니다.
총 파라미터, MoE 활성 파라미터, 양자화 용량, 런타임 지원, 실측 성능을 구분합니다.
- 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 다운로드 전 확인
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.