One DGX Spark, one ~180B MoE at 35 tok/s: 10k lines of code in 8 hours, fully local
Character-Result-281 · reddit · 2026-09-19
A developer built a project end-to-end on a single NVIDIA DGX Spark running Qwen3.8-Flash-Next (NVFP4, 262K context) fully locally:
- 8 hours of planning, coding and testing; 10k lines generated, 800k tokens consumed
- Stack: VSCode Copilot in autopilot mode + SGLang for inference
- The 180B MoE runs at 35 tok/s on the single box
The author admits it's not frontier-model quality, but demonstrates that local large-MoE full-stack development is now practical on workstation hardware.
More from coding & agent
- Butterfly traces multi-agent failures with graphs, 62.5% accuracy vs 53.8% for log prompting — JeremyCMorgan · 2026-11-04
- Free course covers the 4 hard problems of production generative media agents — Al_Grigor · 2026-11-04
- Quota reset night: Codex back to 100% while Claude shows just 10% left — rudrank · 2026-11-04
- Self-improving software is inevitable: closing the agent-observability loop — bibryam · 2026-11-04
- Building a 7-GPU 132GB VRAM local inference rig: consolidate on EPYC or split across two systems? — infectiousstupidity · 2026-11-04
- Founder of 50-person, 12-country team says he talks to AI agents more than humans — Scobleizer · 2026-11-04