VRAM (video RAM) is the memory attached directly to a graphics card. The GPU can only work at full speed on data held in its own VRAM, so VRAM size caps how large a texture set or AI model it can handle, and VRAM bandwidth caps how fast it can stream that data.
VRAM for local AI
A language model's weights must sit in memory while it generates text. A 4-bit quantized 8B model needs roughly 6 GB; a 70B model needs over 40 GB. See how much VRAM for LLMs for a full table.
VRAM for gaming
Games use VRAM for textures, frame buffers and ray-tracing data. Higher resolutions and texture settings need more. Game requirements vary by title; check each game's published system requirements.
VRAM on current GPUs
VRAM and bus width from NVIDIA spec pages; bandwidth from Wikipedia series tables (not listed on NVIDIA pages); MSRPs from NVIDIA press releases. Current street prices: TBD.
Card-specific pages: RTX 5060, RTX 5070, RTX 5080. Not the same as system memory: see VRAM vs RAM.