Nereden başlıyorsunuz?
Sahip olduğunuz donanımdan veya hedef modellerden başlayın; kurulumdan önce kapasiteyi, performansı ve kanıtı karşılaştırın.
Select hardware to see model fit results.
Guides & methodology
Browse the complete catalogs and estimation method when you need them.
Tüm donanım21
Her sınıfın farklı bütçesi, yolu ve gerçekçi üst sınırı vardır.
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.
Rehberi açMac 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.
Rehberi açMacBook 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.
Rehberi açFramework 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.
Rehberi açRadeon 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.
Rehberi açNVIDIA 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.
Rehberi açMac 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.
Rehberi açMac 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.
Rehberi açGeForce 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.
Rehberi açGeForce 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.
Rehberi aç8 GB CPU laptop
Small text models with short context. Expect patient, private inference rather than speed.
Rehberi aç16 GB CPU desktop
Comfortable with 3B–8B Q4 models; 12B is possible only with reduced context and patience.
Rehberi açApple silicon · 16 GB
A polished 4B–8B local experience when the OS and apps have enough headroom.
Rehberi açNVIDIA · 8 GB
The mainstream 4B–8B tier. Some 12B INT4 builds fit tightly with modest context.
Rehberi açNVIDIA · 12 GB
Strong 8B–14B Q4 territory for a single user.
Rehberi açNVIDIA · 16 GB
14B models are comfortable; 20B-class low-bit MoE models are a tight upper edge.
Rehberi açApple silicon · 32 GB
24B–32B Q4 models become practical for one local user.
Rehberi açNVIDIA · 24 GB
The consumer sweet spot for 27B and 30B-A3B INT4/Q4 models.
Rehberi açApple silicon · 64 GB
32B models with long context or 70B Q4 with conservative headroom.
Rehberi açWorkstation · 48 GB
70B Q4 serving begins here, with limited room for cache and concurrency.
Rehberi açDatacenter · 80 GB
Large MoE models and high-throughput servers, still bounded by quantization and cache.
Rehberi açTüm modeller24
Qwen3.8 Flash Next
English- Tahmini bellek
- ~108 GB
- Bağlam
- 262,144 native; 1M extended
Ling-3.0-flash
English- Tahmini bellek
- ~74.4 GB
- Bağlam
- Check selected runtime and checkpoint
Nex-N2.5-mini
English- Tahmini bellek
- ~21 GB
- Bağlam
- 262,144 (config max_position_embeddings)
Qwen3.6 27B
- Tahmini bellek
- ~18.5 GB
- Bağlam
- 262K native
Qwen3.8 27B
- Tahmini bellek
- ~19.5 GB
- Bağlam
- 262K native
Qwen3.8 Flash Next REAP-288
- Tahmini bellek
- ~40.57 GB
- Bağlam
- 7.6K recommended on measured 48 GB Mac
Qwen3.6 35B-A3B
- Tahmini bellek
- ~25 GB
- Bağlam
- 262K native
DeepSeek V4 Flash
- Tahmini bellek
- ~176 GB
- Bağlam
- 1M native
Gemma 3 1B
- Tahmini bellek
- ~1.4 GB
- Bağlam
- 32K
Llama 3.2 3B
- Tahmini bellek
- ~2.8 GB
- Bağlam
- 128K
Qwen3 4B
- Tahmini bellek
- ~3.6 GB
- Bağlam
- 32K+
Gemma 3 4B
- Tahmini bellek
- ~4.2 GB
- Bağlam
- 128K
Qwen3 8B
- Tahmini bellek
- ~6.8 GB
- Bağlam
- 32K+
Gemma 3 12B
- Tahmini bellek
- ~9.4 GB
- Bağlam
- 128K
Qwen3 14B
- Tahmini bellek
- ~11.2 GB
- Bağlam
- 32K+
gpt-oss-20b
- Tahmini bellek
- ~16 GB
- Bağlam
- 128K
Devstral Small 2 24B
- Tahmini bellek
- ~18.5 GB
- Bağlam
- 256K
Gemma 3 27B
- Tahmini bellek
- ~20.5 GB
- Bağlam
- 128K
Qwen3 30B-A3B
- Tahmini bellek
- ~22.5 GB
- Bağlam
- 32K+
Qwen3 32B
- Tahmini bellek
- ~24.5 GB
- Bağlam
- 32K+
Llama 3.1 70B
- Tahmini bellek
- ~46 GB
- Bağlam
- 128K
Llama 4 Scout
- Tahmini bellek
- ~78 GB
- Bağlam
- 10M advertised
gpt-oss-120b
- Tahmini bellek
- ~80 GB
- Bağlam
- 128K
Qwen-Image 2.1
English- Tahmini bellek
- ~33.1 GB
- Bağlam
- Not a token context: one prompt plus up to 10 reference images
Doğru çıkarım katmanını seç15
Ollama
One-command local chat and app integration
Rehberi açllama.cpp
GGUF models, CPU/GPU offload, embedded and unusual hardware
Rehberi açLM Studio
Discovering, downloading and testing models without a terminal
Rehberi açMLX LM
Native Apple silicon inference, experimentation and fine-tuning
Rehberi açoMLX
Apple-silicon serving with continuous batching, memory guards, and supported SSD-backed paths
Rehberi açvLLM
Linux GPU servers, concurrency and OpenAI-compatible production APIs
Rehberi açTensorRT-LLM
Maximum NVIDIA throughput after engine tuning
Rehberi açSGLang
Multi-request GPU serving with prefix caching and speculative decoding
Rehberi açLocalAI
One OpenAI-compatible endpoint for text, vision, speech and image models, GPU optional
Rehberi açLemonade
Ryzen AI, Radeon and Strix Halo PCs, including the XDNA2 NPU, behind one local API
Rehberi açOpenVINO Model Server
Serving OpenVINO IR models on Intel CPUs, Arc GPUs and NPUs with one API
Rehberi açKoboldCpp
One-file GGUF chat on Windows, macOS or Linux with no Python install
Rehberi açllamafile
Distributing a whole GGUF model as one runnable executable on Linux, macOS, Windows and BSD
Rehberi açKTransformers
Very large MoE models with hot experts on the GPU and cold experts in system RAM
Rehberi açUnsloth
Running GGUF or MLX models from a desktop app that also trains and exports them
Rehberi açNasıl tahmin ediyoruz
Belgeli boyut ve nicemlemeden başlar, sistem için pay bırakır ve önbelleği ek maliyet sayarız. MoE toplam parametreyle hesaplanır.
Toplam parametre, aktif MoE, nicemlenmiş boyut, çalışma desteği ve ölçümü ayrı tutarız.
- 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
40 GB indirmeden önce
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.