Inferencing at 10.33 t/s on Qwen 3.5 35B on a $300 laptop

A Reddit user ran Qwen 3.5 35B Q4_K_S on a $300 Lenovo Ideapad Slim 3i and reported 10.33 t/s inference using ik_llama.cpp with two pinned CPU cores, MTP speculative decoding, 64 batch size, and Q8_0 KV cache.

Why it matters: HKR-H/K/R all pass, with a concrete first-person benchmark. Reddit single-post sourcing and limited reproducibility details keep it at the lower featured threshold.