Hardware & ecosystems
What a quantized model is, and why local runs use one
Decode names like Q4_K_M and Q8_0, see where the VRAM and speed savings come from, what they cost in quality, and which level to pick.
Local LLM guides
Choose a GPU, work out memory requirements and compare the cost of running models locally.
All guides
11 guides- Choosing GPUs & running models
Local video GPU upgrade: VRAM or speed
Work out whether your workflow fails to finish or merely runs slow, then use current used prices to decide between VRAM, compute, or both.
- Hardware & ecosystems
Not enough VRAM: does adding RAM help?
Understand what each of three offloading methods moves, and use this site's formulas to calculate how much speed extra RAM can deliver.
- Hardware & ecosystems
AMD GPUs for local LLMs: support and choices
Evaluate framework support, memory cost and setup trade-offs.
- Cost and savings
Budget GPUs for local LLMs by VRAM tier
Compare current listing prices by memory capacity and check hardware limitations.
- Choosing GPUs & running models
DeepSeek GPU and VRAM requirements
Compare local DeepSeek configurations by model size, memory and budget.
- Cost and savings
Local LLMs vs APIs: a three-year cost guide
Compare hardware, electricity and API usage to estimate your total cost.
- Hardware & ecosystems
Mac unified memory vs discrete GPUs
Compare model capacity, bandwidth, cost and software support.
- Multi-GPU and advanced
Multi-GPU LLMs: memory and setup
Understand memory splitting, power requirements and inference setup.
- Choosing GPUs & running models
RTX 3090 vs RTX 5070 Ti for local LLMs
Compare memory, estimated inference speed, used prices and use cases.
- Choosing GPUs & running models
What LLMs fit in 16GB of VRAM?
Compare models, quantization and context lengths for a 16GB GPU.
