Spec tables

The raw data every VRAMwise calculator reads from. Each table is also available as machine-readable JSON, linked inline.

• GPU VRAM & Memory Bandwidth Table

  VRAM capacity and peak bandwidth for consumer NVIDIA, AMD, and Intel GPUs.

• Model VRAM Requirements Table

  GGUF file sizes and VRAM needed for popular local LLMs at Q4/Q5/Q8 quantization.

• Minimum GPU per Model Size

  Quick lookup: minimum VRAM or unified memory per local LLM size class, 8B to 120B.

• Apple Silicon Memory Bandwidth Table

  Unified memory options and peak bandwidth for every Apple M-series chip, M1 to M5 Max.

• Measured LLM Tokens/sec by GPU

  Measured llama.cpp generation speed across GPUs and model sizes — not a theoretical estimate.

Home · Tools · Guides · Methodology