Strata adds Qwen 3.6 35B support, but decode speed loses to llama.cpp on MI50
JsThiago5 · reddit · 2026-10-24
A Reddit user tested Strata's newly added Qwen 3.6 35B support on an AMD MI50 (gfx906, 16GB) and found its decode speed ran worse than llama.cpp. A concrete data point for anyone doing local inference on older AMD GPUs.
More from Infra
- llama.cpp PR drops kq_mask for QSA selection, saving gigabytes of VRAM on Qwen Flash Next — jacek2023 · 2026-10-24
- Chronos: a Postgres-compatible database where agent changes land only if they pass your policies — Antique-Willow-5841 · 2026-10-24
- Sam Altman: OpenAI bets on robots building robots and data centers building data centers — rohanpaul_ai · 2026-10-24
- Full 510GB DeepSeek-V4.1-Flash on a 64GB RAM + RTX 3060 PC: 7.4 tok/s and 98.3% agreement — Muted_Context_1999 · 2026-10-24
- Cloudflare launches Origin Range Requests, cutting R2 p99 TTFB by half — Cloudflare Blog · 2026-10-24
- Benchmarking 7 Qwen3.8 FN inference engines on a single RTX 5090: Basalt and Infernix lead — jesdga95 · 2026-10-24