Browse models by memory size

This page was written and translated automatically by AI and checked by software against its sources, with no human editing. Specifications come from the sources listed; memory requirements and speeds are estimates calculated with published formulas. How we use AI

How many models fit on a generic consumer GPU of each memory size (CUDA run profile, default configuration) at a context length of 8,192 tokens. Only sizes that consumer NVIDIA GPUs in the database have are listed.

Values updated Oct 19, 2026

• 8 GB — 137 models fit
• 12 GB — 223 models fit
• 16 GB — 260 models fit
• 24 GB — 305 models fit
• 32 GB — 311 models fit

Sources

These results use the CUDA run profile; its reserve comes from memory figures in these public llama.cpp logs. Methodology

• ggml-org/llama.cpp/issues/29811
• ggml-org/llama.cpp/issues/27680
• ggml-org/llama.cpp/issues/23679
• ggml-org/llama.cpp/issues/24608
• ggml-org/llama.cpp/issues/23472
• ggml-org/llama.cpp/issues/17931
• ggml-org/llama.cpp/issues/18181
• ggml-org/llama.cpp/issues/20308
• ggml-org/llama.cpp/issues/24758
• ggml-org/llama.cpp/issues/21984
• ggml-org/llama.cpp/issues/27892
• ggml-org/llama.cpp/issues/28167
• ggml-org/llama.cpp/issues/25594
• ggml-org/llama.cpp/issues/20761
• ggml-org/llama.cpp/pull/22546