Running Qwen 3.8 Flash Next With 130k Context on 16GB VRAM at 15-20 t/s
AvidCyclist250 · reddit · 2026-09-24
After a week of testing, the author shares a full recipe for running Qwen 3.8 Flash Next on an RTX 4080 (16GB VRAM) + 64GB RAM, hitting 15-20 t/s at 130k context.
Four things most people miss: the right quant (AtomicChat AD-4.27bpw / Q4KM target with the shared Unsloth MTP head), the right branch (their pr-mtp-fix branch, i.e. llama.cpp PR #28243 plus one fix commit — plain master can't load this MTP head yet), and the right cache flags (q8 KV cache).
The key trick is --spec-draft-cpu-moe: speculative draft experts live in RAM so the target model's hot experts stay on GPU. Versus 10 t/s with IQ4XS, this setup reaches 16.5 t/s generation / 350 t/s prefill at 131k context. Details in the GitHub repo.
More from Infra
- UK Sovereign AI invests in Fractile to build next-gen frontier AI chips — charlieharris01 · 2026-11-11
- What happened when I moved vector retrieval into Snowflake SQL instead of orchestrating RAG outside — dima806_dima · 2026-11-11
- AI chip startup Fractile raises $720M Series C for DRAM-on-processor design — Tom_Westgarth15 · 2026-11-11
- Mobile AI Factory: agentic pipeline builds on-device SLMs that beat 5-shot frontier models — kalyan_kpl · 2026-11-11
- Vivo's Starlink direct-to-cell service hits 10M users in two weeks across Brazil — XFreeze · 2026-11-11
- Nan Ya Plastics to spend NT$11.53B on fifth glass fabric plant on AI server demand — zephyr_z9 · 2026-11-11