llama.cpp Adds GPU Cache for MoE Experts in Host Memory

Developer am17an submitted PR #29887 to llama.cpp adding a GPU cache for MoE experts kept in host memory, potentially delivering big speedups for Mixture-of-Experts models that don't fully fit in VRAM.

• llama.cpp merges MTP support for Qwen Flash Next2026-10-01
• Four open decision models benchmarked on an RTX 40902026-10-08
• User builds 256GB VRAM rig with 8 Radeon Pro V620s2026-10-08
• llama.cpp adds local support for GLM 5 Flash MTP2026-10-08
• llama.cpp takes the stage2026-10-07