Can I run an LLM without a GPU?
Yes, but large models will be slow. Tools such as llama.cpp run on ordinary CPUs, and speed is limited by system memory bandwidth, about 90 GB/s for a typical dual-channel DDR5 desktop. Small models (7 to 14 billion parameters at 4-bit) are usable; 70B models crawl.
Apple silicon Macs are the exception that makes this question less relevant: their GPU is built in and shares all of the system memory, at 307 GB/s on an M5 Pro and 1.2 TB/s on an M5 Ultra.
Mixture-of-experts models with few active parameters, such as gpt-oss-20b, run better on CPUs than dense models of the same size.
Sources
Updated