Build the rig around the workload
Practical guides for memory, quantization, platform choice, and total system cost.
How to buy hardware for local AI
Start from model size, quantization, context, and runtime—not GPU marketing.
Inspect →GUIDEVRAM vs RAM vs unified memory
Understand dedicated VRAM, system RAM, Apple unified memory, and offload.
Inspect →GUIDEQ4 vs Q5 vs Q8 vs FP16
Compare memory footprint, quality trade-offs, and runtime compatibility.
Inspect →GUIDEWhy context length changes VRAM
KV cache grows with context, batch size, layers, and architecture.
Inspect →GUIDEWhat AI models fit 12 GB VRAM?
Map realistic small and mid-size Q4 options.
Inspect →GUIDEWhat AI models fit 16 GB VRAM?
Understand the popular 16 GB hardware ceiling.
Inspect →GUIDEWhat AI models fit 24 GB VRAM?
Explore the RTX 3090 and RTX 4090 capacity class.
Inspect →GUIDEWhat AI models fit 32 GB VRAM?
Plan around the RTX 5090 memory ceiling.
Inspect →GUIDEWhat AI models fit 48 GB VRAM?
Workstation and data-center options for larger models.
Inspect →GUIDEApple Silicon for local AI
Unified memory expands capacity but runtime and bandwidth still matter.
Inspect →GUIDEAMD GPUs for local AI
Capacity can be attractive; verify ROCm, OS, and runtime support.
Inspect →GUIDEBuying a used GPU for local AI
Inspect memory, thermals, power, warranty, and total system cost.
Inspect →GUIDEMulti-GPU local AI rigs
Model sharding adds capacity but also complexity and bandwidth constraints.
Inspect →GUIDECPU and RAM offload explained
Run larger models with slower token generation and more system memory.
Inspect →GUIDELocal GPU vs cloud GPU
Compare upfront cost, utilization, privacy, maintenance, and hourly spend.
Inspect →GUIDEInference hardware vs training hardware
A system that runs a model may be unsuitable for fine-tuning or training.
Inspect →GUIDEHow much SSD storage for local AI?
Plan for multiple quantizations, caches, datasets, and model updates.
Inspect →GUIDEPower and cooling for AI workstations
High-memory GPUs also bring PSU, heat, noise, and chassis requirements.
Inspect →GUIDEMini PCs for local AI
Unified-memory mini PCs trade compactness for software and bandwidth limits.
Inspect →GUIDESeven VRAM buying mistakes
Avoid buying speed without capacity or capacity without software support.
Inspect →