8 Coding Agents on One RTX PRO 6000: Measured Throughput for Local Qwen Deployment
AIFlow_ML · x · 2026-11-04
Local AI Daily (Nov 4, 2026) aggregates measured local-AI runs: hyudryu published a full pinned recipe for running NVIDIA's Qwen3.8-Flash-Next-FP4-Local on a single RTX PRO 6000 with vLLM V2 (b12x kernels), FP8 KV cache, 262K context, depth-4 MTP drafts and a 450W cap. Coding-workload measurements (temp 0, thinking off) hit 257.4→287.0 tok/s for one user and 669.5+ tok/s total with 4 concurrent users, with Docker builds and raw JSON in the repo. Also covered: running DeepSeek across two Macs. The newsletter labels data as MEASURED / VENDOR-CLAIM / ESTIMATE.
More from coding & agent
- Game dev pushes back on 'full game in 24 hours with AI' claim: it's a demo, not a game — AIandDesign · 2026-11-04
- YC-backed twentyfour26 builds AI machinists that plan and run CNC machines — ycombinator · 2026-11-04
- Shopify CEO hails Tomo agent buying via Shop Pay over iMessage — Jackyhuang · 2026-11-04
- Stripe hires engineers to crack hard problems in agentic commerce — jeff_weinstein · 2026-11-04
- xAI engineer yacineMTB: agents work autonomously while I sleep, I wake up to breakthroughs — yacineMTB · 2026-11-04
- A start-to-finish workflow for AI dialogue scenes, from script to 1080p — DavidmComfort · 2026-11-04