Developer weighs EPYC Milan and 2–3 RTX 3090s for self-hosted coding agents
OneWolverine307 · reddit · 2026-07-29
A developer asks whether an EPYC Milan + Supermicro H12SSL-i + 2–3× RTX 3090 build is still the best value for self-hosted coding agents.
- The target is an Architect→Developer workflow for PHP/Laravel, Python/FastAPI, and Angular/TypeScript.
- The proposed rig centers on 256GB ECC RAM, 48GB of initial VRAM, and room to expand to 72GB VRAM later.
- The post is really about the economics of local inference for serious coding agents: cheaper platforms, Threadripper Pro tradeoffs, and whether more VRAM or more system RAM matters more.
Related event: Developer Evaluates $7,500 Local Coding Agent Workstation(2 posts)→
More from coding & agent
- Agent debugging gets easier when you separate loops, graphs, and harnesses — Roger_M_Taylor · 2026-07-29
- An AI agent found a real bug while automating Apple Ads changes — charlieholtz · 2026-07-29
- Multi-agent coordination failures are still hard to study without real production traces — Icy-Weakness8310 · 2026-07-29
- A multi-model subagent workflow uses Grok, Kimi and DeepSeek to compile AI news — op7418 · 2026-07-29
- Kimi K3’s 2.8T open-weight model is built for supernodes, not Mac minis — APPSO · 2026-07-29
- CodePilot v0.61.0 adds Opus 5, Sonnet 5, and Grok 4.5 sub-agents — op7418 · 2026-07-29