Interesting - in my setup (llama.cpp rtx5090 qwen-3.6 27b) prompt processing with mtp is almost half vs non mtp. Sounds like I need to investigate what is wrong.
Maybe enabling MTP causes some weights to be displaced to host memory? MTP itself doesn't do anything during prefill so that should be exactly unchanged, decode will vary depending on settings but with 2-4 proposals depending on workload I've never seen an overall slowdown.
edit: I recommend building recent llama.cpp from source, I've been updating about once a week, as there has been a fair amount of work related to MTP recently. If you're running a lot of tool calling on Qwen you might also benefit from one of the bugfixed chat templates like the Froggeric version.
edit: I recommend building recent llama.cpp from source, I've been updating about once a week, as there has been a fair amount of work related to MTP recently. If you're running a lot of tool calling on Qwen you might also benefit from one of the bugfixed chat templates like the Froggeric version.