Back to news
researchMoonshot AIIntel2026-11-13

Developer Runs 2.8 Trillion-Parameter Kimi K3 on Retired Intel Optane Memory

A developer ran Moonshot AI’s 2.8 trillion-parameter Kimi K3 model with no GPU, using retired Intel Optane persistent memory and llama.cpp, achieving about 1.5 tokens per second at one-bit quantization.

A Reddit user reported running Moonshot AI’s 2.8 trillion-parameter Kimi K3 with no graphics card, using retired Intel Optane persistent memory and llama.cpp. One-bit quantization reached about 1.5 tokens per second after enabling direct access.

Kimi K3本地推理傲腾