← sources

176 t/s decode and 6,450 t/s prefill with Q4 Qwen-3.8 Flash-Next on 2xR9700 + 64GB DDR5 using Deadcode's Radiance engine

source: reddit (reddit:LocalLLaMA)published 2026-10-14 21:44ingested 2026-10-15 02:42

retained text

https://preview.redd.it/kudj7xii5ivh1.png?width=1508&format=png&auto=webp&s=07712e81310b81aceb2cf8f4d8a7c2eadeb0c304

Just built a new PC, full [parts list is here.](https://ca.pcpartpicker.com/list/7pYdGk) I am using Deadcode's[ Radiance inference engine](https://codeberg.org/StillDeadcode/radiance), optimized for 2xR9700 and [his Q4](https://huggingface.co/StillDeadcode/qwen3.8-next-flash-fp8-iq4r-moe). The throughput benchmark tool is [BetterBench.](https://github.com/GGZ14/BetterBench)

document fetched 2026-10-15 00:45 · canonical https://reddit.com/r/LocalLLaMA/comments/1x640u6/176_ts_decode_and_6450_ts_prefill_with_q4_qwen38