tandem.cpp: Qwen3.8-Flash-Next at 97 tok/s on a single 4090 + 64 GB RAM, 3.3× llama.cpp, works for any MoE
tandem.cpp is a patch set for llama.cpp that makes MoE models that don't fit in VRAM run faster. It keeps the most-used experts on the GPU and lets the CPU help with the rest, for any MoE model llama.cpp supports. Single-user decode is the target.