Homelabber runs 2.8-trillion-parameter Kimi K3 on old Optane memory, with fans to stop the RAM melting

A homelab builder on Reddit has run Moonshot AI's 2.8-trillion-parameter Kimi K3 model on discontinued second-generation Intel Optane PMem 200 memory, using a 40-core third-generation Xeon Scalable CPU instead of a GPU, Tom's Hardware reports. Even with 1-bit quantization, the build manages a sluggish 0.85 tokens per second.

The setup runs hot. The builder's advice is to "treat active DIMM cooling as mandatory", and they aimed two 12W blower fans at the modules to keep temperatures under control, warning that Optane's heat can burn out nearby DRAM.

Kimi K3 is not built for home use: Moonshot recommends data-centre inference engines, and even the smallest community quantization needs at least 610GB of RAM. Secondhand Optane is one of the few ways to reach that capacity at home, and with DRAM prices sharply higher, the builder says the Optane memory cost "a fraction" of the equivalent. The project follows a separate user's build earlier this year on first-generation Optane.