2x AMD R9700 hits 176 tok/s decode and 6,450 tok/s prefill on Q4 Qwen3.8-Flash-Next
Athabasco · reddit · 2026-10-15
A Reddit user's new dual AMD Radeon AI PRO R9700 (2x48GB) + 64GB DDR5 build, running Deadcode's open-source Radiance inference engine with a Q4 quant of Qwen3.8-Flash-Next (200K context), benchmarks at 175.8 tok/s combined decode (batch=1), 6,453 tok/s median prefill at 64K tokens, 74ms TTFT P50, and 448.8 tok/s aggregate at 8 concurrent requests. Full parts list, engine and BetterBench tool are public — a reproducible reference for consumer AMD local inference.
More from Infra
- vLLM + NVIDIA wideEP Co-Design Makes GB300 NVL72 7.2x Better Perf per TCO — woosuk_k · 2026-10-15
- Simulating 10.5M scenarios to find the most cost-effective AI chip location verification method — chrisrohlf · 2026-10-15
- Neocloud stocks fall 12% in a day after CoreWeave's bond selloff — SumitGup · 2026-10-15
- TRELLIS.2 image-to-3D runs at full 1536 resolution on a 16 GB card via tiling patches — over-yonder-tech · 2026-10-15
- ServeTheHome tours AWS's secretive networking lab: timing, optics and hyperscale — firstadopter · 2026-10-15
- vLLM adaptive verification now works head-free, using draft logits with online logistic calibration — vllm_project · 2026-10-15